Anand Mundada, Gaurav Tungatkar - Uber [slides]
Anand: we have driver rider and vehicle, you request a trip and magically it drops you off at your destination
<ted> [transportation today]
<ted> various modes of transportation, personal, food deliver, bikes and scooters... freight, fleets
Anand: when looking at
all these customer types we see some key things that need to be
addressed
… privacy and security,
support user and business accounts
… need to consider the
relationships between entities
<ted> [privacy and security]
Anand: for a given
customer you have a lot of potential policies surrounding
authorization
… as noted we have a
number of different customers types, many things in common but
also special considerations
… driver has to have
good ratings, driver license and so on and customer a payment
method
Gaurav: [describes extensible
data model]
… we think of modeling
in three components, uuid, base schema and extensions
Dick: does Uber rate their riders? I know I can rate the riders
Gaurav: Uber does not...
Jay: @@
Carlos: how do you manage versioning?
Gaurav: we always have
to be forward and backward compatible
...we eliminate breaking
backwards compatibilities
George: slide you had
wrt model for relationships with names and properties
...at the graph workshop
there was question whether you should have metadata on the
relationship or not
...you have a specific
query language?
Gaurav: not right now
DavidM: there is how
you can address extensible, how are folks suppose to go back
and check for evolution of use cases
...how do you add
properties to those edges without conflating and making sure
the data is accurate for associations
Gaurav: what we are mainly concerned about are the data models themselves whereas you're talking about infrastructure, os or bugs or misuse of data
David: misuse, spoofing can occur at sensors and many layers
Gaurav: I am sure we
have people looking at information security but have not
thought about it from a graph perspective
...we have had
iterations of the customer model itself
Anand: we control when
you add extensions, ensure your use cases warrant a new type of
customer
...we need to know which
are used by what
Julián Rojas - Ghent University [ Slides]
[slide 1]
Julián: we need data, about what, mobility
preferences... how do they want to travel
...what kind of service
subscriptions and accounts they have, do they have a metro
card, uber/lyft, bike share, wheelchair or have other special requirements
...we need to be able to use
but protect this data
... this is where SOLID (Social Linked Data) comes in
... solid allows you to use the apps you need while storing the
data you want
Julián: instead of being in silos like
facebook, linkedin etc; solid it is a decentralized mindset
...web on applications are
often designed with hardcoded links to data sources on their
end, not taking advantage of data on the web
... service requests access to use their data from the
user
... combination of W3C standards are behind SOLID RDF,
WebID-TLS, Linked Data Platform (set of protocols for handling
access)
... important to gain adoption to have some tools for the
developer [tools slide w/ links]
... GraqhQL-LD is a wrapper to translate to SPARQL queries
underneath, results in RDF
... important for traveler profile to have their needs, similar
for a car's capabilities as EV can go into zero emissions zones
for example
... thanks to GDPR, De Lijn a Flemish language transportation
provider is working with us to give the users back their
data
...SOcial LInked Data
(SOLID)
...Data is personal,
should be in the control of the user. SOLID decouples the data
from the logic of the applications.
...Traditional apps are
based on centralization of data. SOLID takes the opposite
approach.
...each person has 1 or
more of their own data 'pts' and they decide to whom access is
given.
...SOLID uses RDF for
data repn / WebID for IDs / WebID-TLS for authentication /
Linked Data Platform for data access interfaces
Julian: we need to describe user
preferences, mobility needs, service subscriptions etc as
ontologies
...plenty of tools
available already including node.js server, readct generator,
angular generator and graphql-ld
Deborah: what sort of identifier
used for an individual. take rental car scenario, what data
about me belongs to them?
... is there a standard notion of it being encapsulated
Julian: unsure of the protocol details but it negotiates what can be [re]used by the rental company
* question from Debra on _how_/what data is used to identify users? is there a standard notion of how my data is tagged? A: AFAIU WebID defines how to handle the uniqe IDs that go inside the pot. Permissions need to be granted first before anything is shared.
<ted> Adnan BMW
<ted> Inrupt, Timbl's startup. He is on sabbatical from MIT & W3C at present. SOLID work continuing at MIT
* q from Anrnad: at BMW we tend to look to wards the GDPR. e.g. renting a vehicle requires reading the agreement because there the rental company captures data basedon that vehicle account.
Ben Gardiner - NMFTA [ Slides]
Ben: our members are freight
carriers, we started interviewing them on needs and the
telematics service providers(TSP) like Geotab
... I have a cybersecurity engineer background but not
breaking stuff at present. we are non-profit, not sales
pitch
... working in HVCS (Heavy Vehicle CyberSecurity)
[Motivation]
Ben: freight deliveries is a very
lean business, data helps them optimize but also required not
by safety regulations
... concern is TSP can go out of business and leave fleet
operator in the lurch
... TSP interaction is not dictated, we are only interested in
how the things are handled in the cloud not how it gets
there
[Portability Use Cases]
Ben: compliance and safety
monitoring, ERODS, quality of data, monitor vehicle status,
etc
... we interviewed our carriers to create these use cases, they
really want all this stuff
[State of health]
Ben: all TSP are required to give to the regulators, what if that goes away - they're out of compliance and need a fallback
RJ: also for roadside inspections
Ben: ELD need to be ready to do
data transfer at the side of the road too
... the fleet operators want to be able to get data export on
daily basis, not necessarily to their infrastructure but
somewhere to put a copy
... driver routing and trailer packing requires coordination,
think of pallets like passengers in a plane (but making
multiple stops and wanting to optimize boarding
groups...)
... fleet operator often needs to be able to send out a
geofenced broadcast to all their drivers
... also being able to respond from drivers
[Compliance and safety monitoring]
[In field Maintenance]
[Estimate on data quality]
Ben: Geotab does a different
sampling method than the rest of the carriers
... need to be able to compare apples to apples
@@harjot
Ben: we learned suprisingly some
wanted the raw signals (heavy vehicle signals actually
standardized unlike auto CAN, SAE J1939)
... they use this to give themselves some portability
[Monitoring Battery Use Cases]
Deborah: does it require authentication to get J1939
Ben: no, some are using a preshared secret and SAE is working on auth
Carlos: what trends and use cases
are you seeing?
... EV and AV are starting to emerge
Ben: EV being used more for local delivery than long hauls at present given recharging time
Carlos: what other metrics besides fuel and battery do they care about?
Ben: fuel optimization is a huge
part
... they need sufficient location history to predict usage and
prebuy gas
Jay: how is this being utilized?
Ben: what we did was work with TSP to understand what they already do with TSP
Rudolf Streif, ibeeto [Slides
Rudi: I am working with clients
to bring proof of concept solutions to production and
consulting W3C
tr
... I was a consultanting system architect with JLR for 5
years, heavily involved and chaired activities at W3C and
GENIVI
... W3C Auto WG about developing specs for open web platform.
we created Vehicle Information Service Specification
(VISS)
... challenge is how to provide vehicle status in a consistent
way across manufacturers without having to reverse engineer
OBDII for each manufacturer
... also varies by model over the years or even change with
software updates over the years complicating this further
... JLR wanted to solve this problem and be able to create apps
to run on the head unit without having to rework them
... convential approach is a "Fat API", the problem is every
time you add new signals you need to update the spec for
addition methods which is tedious
...The w3c auto group
missign is to dev specss for an open web platform for
developers to get access to vehicle data via IVI systems and
vehicle data protocols
...many mnay
participants from multiple OEMs and suppliers
...The challenges with
creating the VSS include: OEMs want to extend an API, security
mechanism are required that provide authentication adn
authorization, application devs want to have a way to rapidly
deploy apps across as many vehicles as possible
...The typical 'fat' API
approach of a getting data values via named functions comes
with the problem that every time you update the set of data
provided you also need to update the specification of the API
via it's function names
...VSS takes a different
approach
Rudi: we looked at doing
something different, Vehicle Service Specification is the data
model at GENIVI which we settled on
... for the VISS API we only needed a handful of methods to
expose the signal tree
[Vehicle Signal Tree]
Rudi:Each part of the vehicle is repn'd as a node. The signals and controls are leaves on these nodes. The nodes can have multiple instantiations (e.g. 4 seats in a car)
Rudi: green parts, branches, blue
nodes the leaves which may be a sensor or actuator, red the
attributes
...Wildcards allow
access to entire sets of signals.
...The structure is
'discoverable' by developers who aren't necessarily
indoctrinated into the usual automotive cryptic
nomenclature
Rudi: we followed Keep it Simple and provide developers more intuitive
Josh: those relationships have semantics?
Rudi: not really
Ted: @@BMW/EU...
Carlos: you have average consumption, why is that in the taxonomy?
Adnan: you want to make it
available instead of requiring each making it from
scratch
... we should note we have a pending pull request for VSS2.0
that reworks things somewhat
* q from Carlos: why do we have to moel something like 'average' in the hierarchy. That is an aggregation of a signal. Why is it included in the taxonomy. Adnand andwers: this singal is actualy by calculated on the behicle. They decided that since it is beging calculated already tat it should be forwarded on to consumers. Also this is v1 and would like to improve.
[Addressing example]
[Spec format[
Rudi: spec is in YAML
... min, max, unit...
...The signals in this
tree can be addressed via a dot-notation. Leading path
components represent the brnaches. Last component is a leaf
...the specification
format is YAML lists. includes data type, unit min/max and
comments
Rudi: we have parsers to generate additional formats JSON, Franca, CSV...
https://github.com/GENIVI/vehicle_signal_specification/
[Vehicle Data Interfaces Architecture]
[VISS Overview]
Rudi: the specification
format is YAML lists. includes data type, unit min/max and
comments
...The repo includes
tools to generate other specification formats. Following GENIVI
project model.
...The repo uses
'typical' github fork/branching flow
...This data model is
used in both the IVI headunit runtime and also REST interfaces
exposed as internet-facing services.
[Service Messages]
...Also a websocket
based server on-vehicle. Javascript libraries are made
available for easy integration.
...The API include
authorize cmmand to implement access control . Request are made
with security token which are later used to gain access to
trees of signals (conditional on permissions granted to the
tokens)
...Plans for future work
include hosting a micro-service REST server on-vehicle for
easier integration with HTML5 based apps in IVIs
github.io/w3c/automotive
* an attendee at the back notes that there is a reference impl available also
[Slides]
Ted:
W3C is doing
automotive standards to: make a robust app ecosystem, increase
interoperability, enable use of the much larger web-devloper
talent pool
...still working on the
rich ecosystem target ; we are currently focused on signals
...The VISS (as distinct
from the VSS) was created for telematics: notifications, media,
LBS/Nav, Payments, other under construction
...future specs will
enbale taking data and services external to the car and making
them available to apps in the car
...There is so many
examples of data-driven decision in vehicles -- we need to
normalize and standardize so that we can each leverage work and
not each re-implement our own solutions
<Harjot> re OTAPI presentation, 4 types of engine data for ELD logs: Engine Hours, Odometer, Engine Based Road Speed, VIN
<ben> thanks Harjot!
q ben: what is the relationship btw VISS and VSS. * VISS is the API interface which uses VSS.
* q from David M. maybe there should be work on diagnostics of data quality? maybe work on integrity of sensor signals (e.g. cypher physical security)? A: we hare having those discussios in out task force on data
<Harjot> Q: Difference between VSS and VISS | A: VSS (vehicle signal specification) is the data model for how the information is structured (trees / node hierarchy). VISS (Vehicle Information Service Specification) consists of the lightweight APIs used to access data from (complex) data models. Links: https://github.com/GENIVI/vehicle_signal_specification & https://www.w3.org/TR/vehicle-information-service/
Adnan Bekan - BMW
Adnan: for IoT we wanted to think
of creating different experiences for our drivers
... we did a map of what is existing, quite a few different
standards efforts and solutions
... to come closer we needed an ontology to create a W3C Web of
Things (WoT) thing description
... VSS was a great data model for us to build an ontology on
called VSSo
... we build a setup with node-red, originally developed by
IBM. it was perfect for our demo
[filled out WoT thing description]
[flow example]
[video of demo]
Adnan: we had a smartphone device
sending position of vehicle, it could auto lock if you went 50m
away for instance
...They have built a VSS
ontology, VSSo. covering SOSA and VoT
...They also implemented
a PoC integrated with a BMW i3
...built a flow in
Node-RED. intrgrated a B<W i3 with a node-red flow to do
e.g. unlock/lock, blinking lights etc.
...then they a feature:
when the user is far from the unlocked vehicle, the vehicle
lock is activate.
...shows a video of the
demoy
Josh: can any BMW owner use
this?
...q from Josh: is this
something that any BMW owner could do? A: no. we are not
opening up the API. You can do this via some reverse
engineering.
Adnan: no, we are not supporting that nor exposing API at this point
Dick: have you thought about the security aspects
Ted: its a demo, of course they are thinking about security before making it available
Adnan: yes, there are ways we can authenticate device, owner etc...
* q from Dick R: this is a cool demo, have you planned out how to do auth? a: we ave a security team , with the VoT you do get the ability to use stack elements like OAuth.
Lothar Determann - Baker McKenzie [ Slides]
Lothar: if you are interested in
attendeing the World Economic Forum on October 2nd
... some of the points are already well know for this
audience
... there is a battle for data access going on right now
... data exists in physical realm, our minds, on hard
drives..
... the definition for personal data is pretty broad
... from a privacy perspective mindset in US goes back to the
4th ammendment
...The california
privacy act of 2018 takes effect in Jan 2020. Applies to any
information that can be linked to a person or household.
...speaker asserts that
pretty much anything can be linked to a person or to a
household
...(generally)
no one really has ability to solely claim ownership of
data.
...data created by these systems is not owned. in the sense of
property law, you cannot prevent access to it. Lawmakers are
carving out ownership from data privacy laws.
...speakers thinks that
the genie is out of the bottle. data minimization will not
solve the problems. neither will data ownership
...minimum
periods of data retention -- is an inaffective (primary) method
to faciliate data privacy
...the genie is out of the bottle. data minimization will not
solve the problems. neither will data ownership
...all of the concepts
in data protection are separate: privacy / data security / data
privacy / data protection / data ownership / data retention /
data residency / trade secret protection. They often get mixed
up for advocacy, political reasons
...we don't want to
bring concepts from ownership of property into data; we don't
want anyone to own info that is in our head which is ultimately
where information resides
...trend predictions:
restriction on data sharing, selling (impact on competition,
innovation) / data protection as trade measure / data residency
/ data security threats / privacy law enforcement
...data
retention and data residency laws are contrary to the notion of
data privacy. Ensure access to data by regimes
..."I am not a very
private person despite dealing in privacy issues"
Martin Kurze - Deutsche Telekom [ Slides]
[EU Project, partners, credits]
Martin: ULD, lawyers are supportings this project
[GDPR - Global Data Business Preventer]
Martin: not as bad as it
seems...
... what is allowed, how do you make things legal
...GDPR is 'not as bad
as it might seem' -- the basic rule is everything is forbiddne
but then there are several hundred pages on what is
allowed.
* speaker thanks the lawyers b/c the laws are there to product people and in this case, their data
"for a specific purpose" is an important aspect
[SPECIAL Objectives]
Martin: transparency is easier than enforcement
[scalable policy aware linked data architecture]
Dick: I noticed you didn't
mention ownership but control
... can you help me understand what that distinction is?
Martin: there is data related to
you largely not owned by you but you should be able to have
some control over it
... granular definition, what data to which parties for what
purposes
... clear transportation has sizable data needs and it includes
personal data
...the project, SPECIAL,
objectives include a policy management framework. givens users
control of their data / a transparency framework / a scalable
linked data archticture / a pilot implementation /
collab+dissemination+standardization
* q from Dick R (long) about ownership, licensing, control... A Martin: we have policies, we can control, etc. A Lothar: this is a non-issue. people can sell what they do not own. e.g. in china people sell fresh air from canada, they do not own the air in canada.
[Special approach: policy language, vocabulary and policy engine]
Martin: policies are in theory enforced during interaction but in actuality that is difficult
[Telecommunications perspective]
Martin: with 5G we will have an
opportunity to collect more data but also increased
responsibility
...the approach in the
SPECIAL project: take as inputs: payload data, permission,
policies. Policies are persisted along with the data. Querying
exposed by APIS is provided in a policy-aware way.
...DT imagines that it
will fit into tomorrow's world of transportation by being a
trusted 3rd party and a service provider
Martin: this group is probably aware of W3C ODRL (policy language)
Mark: cell provider can track
consider amount of information and correlate that with other
data to basically understand all your actions throughout the
day
... we leave digital traces everywhere. there is recognition of
control, ownership etc but not anything on de-anonymizing and
usage independent of control
... the CA law is interesting on a case by case basis
... we are thinking less about the purposeful use of the
data
... I understand how this technology can help us a great
deal
Martin: actually DT doesn't sell data because they are restricted which is why they are interested in these sorts of projects to be able to responsibly
* q from Mark: given the existnce of businesses that exist to de-anonymize based on cell phone data, how will these policies effectively protect people's private data? how do we deal with this reality? A: 1) DT is almost bankrupt because they do not sell the data 2) transparency and trust is key with this approach you give the user control and visibility into their data and its uses
Lothar: I am against massive restrictions. there is adequate harm based law to leverage instead
* the lawyer weighs in: in the US there are harm-based privacy laws. So consequences can happen for cases where harm is found to be caused by data sharing
Arjun Hassard, NuCypher [ Slides]
[practical privacy]
Arjun: privacy capabilities
becomes important when combined with presentation
... what this boils to is access control
... I'll explain cryptographic protection
...Trust minimizing
tooling for the automotive sector
...pactical privacy
defn: do useful & typical things with data e.g. m->m
sharing
...privacy-preservation
defn: preserve encrypted state or avoid unnecessary
decryption
...it all boils down to
enforcing access control. clasical approach is trusted third
parties.
...nucipher approach is
threhold proxy re-encryption & decentralization
Arjun: let's work through a
sharing scenario
... alice's data takes decryped into it is key encrypted and
share with bob and charlie
... to encrypt to share with a specific individual you need
them to be online to get their key
...why not pubkey
crypto? the problem happens when data needs to be re-shared
with other recipients. Requires decrypt-then-encrypt. Where the
decrypt happens could mean privacy is not preserved
[How proxy re-encryption works]
[animated example]
Arjun: you can include arbitrary
conditions (eg payment)
... if you distribute/decentralize this architecture you
protect against ddos, outages etc
Rudi: how does the re-encryption take place w/o Alice's private key
Arjun: we have a lengthy white
paper explaining in more detail
... this is policy driven and Alice can decide which
individual[s] or organizations to share with
[interaction diagram]
Arjun: Alice doesn't need to
negotiate access to Bob's public key
... you can apply this to many use cases in transportation
[slide with high level use cases]
...proxy re-encryption.
Alice shares a re-encryption key with a server, the server
performs re-cryption with Alice's permission when given Bob's
public key. Furthermore, nucipher uses split keys to prevent
denial-of-reencryption-service by a single proxy.
...scheme name
umbral
...in automotive
application: alic give public key to vehicles OBD device.
Device stores data encrypted with alice's public key. Alice
creates a policy and gives acces to the insurance company.
Insurance company gets encrpted data and request re-encryption
to them from nucypher network.
...more applications
than just insurance
Arjun: who creates the sharing
policy, where is it enforced - on the car or cloud?
... this raises the question at what point does granular access
become a burden to the customer?
... without becoming a UX disaster
David: is your tokenization dependent on etherium blockchain?
Arjun: some metadata that ends up on chain
[discussion of economics of sharing proxies]
Rudi: this is related to what
media is doing to protect media streams
... entitlements are handed to subscribers
vehicle can write to policy but private key remains with Alice
before going for re-encryption
Mark Fox - University of Toronto [ Slides]
Mark: ISO JTC WG11, mandate
includes data ontologies for smart cities
... challenge is access to the specs or ontologies in ISO
... I want to have an open process and help me develop what
that is
... we need something openly for the data world
... this is a set of proposed new work projects, focus on data
models
... not limited to transportation per se
... from yesterday you remember the model levels we are doing
for city data. foundational (time, location...), city and
service
... we want to identify those common concepts, what they can
contribute to
... at service level, they are created by one but available to
other services
... we are not trying to reach down to individuals vehicles but
stay at a higher level
... it is about commonality
David: is this clustering (aggregation)?
Mark: that could be the case and
certain types of information may make it way out towards
us
... how do we do this?
... appreciate W3C bringing us together for this
conversation
... my goal is to not create a closed process but open
one
... what I am proposing is to use an open tool, not necessarily
github, so they are openly available to see, people can
register and comment
[Open Development Process]
Mark: want to take a minimalist view, why the properties being proposed are relevant in city scope
Dick: one of the concepts for L2
concept was need for multiple use cases to justify existance,
why?
... others should be able to make a decision
Mark: at service level, go ahead, at city level demonstrate other use cases
Dick: is there a graduation process to move it to next level?
Mark: there is a process for
handling this
... that is the point of an open process is for those types of
things to occur
Deborah: concept of parking space can go from household to office
Mark: we are not trying to do the detailed work taking place at other organizations
[see https://www.w3.org/auto/events/data-ws-2019/schedule]
Grade Crossing Behavior Model RJ Rittmuller
Rudi: you are training on model of good behavior, not bad
RJ: we are
... you can simulate behavior
Open Telematics API, Ben Gardiner
ADEPT – a Limited Iterative Ontological Notation, Gregory Sharp
David Mordecai
[background and coordination with other universities and NYU]
[refer to video in lieu of slides]
Cheryl Scott
[EU data safety project video, see zoom recording]
Inter-SDO Coordination
Standardized Entities for transport
Aftermarket
<jasont> data for road safety video https://www.youtube.com/watch?v=zfNeHtUNRlA
Brent: we focused on aftermarket
such as a more advanced tpms being able to provide additional
signals
... challenges on how to get access to this in the vehicle and
send this to the cloud
... for fuel %of full as in rental car market. they are
inconsistent in practices and even tank size per vehicle
sometimes
... atmospheric pressure and other influences
Josh: we had standardized entity
types
... observed considerable overlap in schemas presented and ours
for things like devices, sensors, users, trips
... thought it would be useful to create a shortlist and we
actually came up with quite a few more
... different perspectives on what a trip is
... a trip can start when it is turn on, duration and route
taken until it is turned off
... need to look at schema.org to see which are already have
notions established to build on
Ken: we talked about how to make
inter-SDO happen
... we modified the pyramid model I showed. worked to merge it
with JTC-1 City concepts
... model is no longer an ITS model but more an IoT model
... encourage industries and other groups to fit that
... it should be divided up into different standards and parts.
ontologies, issue handling etc needs to be public and
considering using github for that
... we need to figure out logically which SDO or entity is the
driver
... this requires us to go back to ISO to see if this is
possible from their perspective
... not detailed modeling rules
Deborah: we talked considerably
about security issues, looking at the gaps, the now and what is
coming down the path
... we are not implementing the known basics, integrity
protection, encryption etc what is being done elsewhere
... manufacturers know this and working toward solution but not
fast enough
... scalable attacks are more pronounced, concern about over
the network attacks in particular
... SOTA as a practice that needs to be used widely and ensure
not provide an attack surface
Notes:
Security breakout: what are the gaps/challenges? Rudy; security in V2X space; Red shirt/shorts: network security; we’ve done this before—solved, not adopted; multiple trust roots RJ: Vendors not implementing standards, e.g., authentication, Android auto —failure to follow reference implementation; lack of technical know how Red shirt: Cost of ECU with hardware security modules RJ: not implementing the basics RJ: some vendors are monitoring their networks—Tesla is monitoring their networks; have a security gateway RJ: ECU failures from exploited vulnerabilities can be catosrtophi Rudy: remote access enables scalable attacks Amir: would not see evidence in telematics data of vehicle controls sent to vehicle RJ: USB-connected cameras watching the freight Amir: Geotab does passive collection; other telematics companies are offering control of the vehicle though the CAN bus Red shirt: need certified dongles Khanh Hoang: Q: does Geotab dongle have built in GPS? A: Amir: no wifi, cellular back end Khanh: Cadillac on super cruise is geofenced Amir: slowing down stolen vehicle Amir: options offered: remote start, lock/unlock RJ: example of Tesla owner giving cross-site scripting attack as name of vehicle; ran when service person opened up vehicle record; could have affected fleet Rudy: too many problems with CA-based models — doesn’t scale/isn’t tractable in vehicle space Challenge: fast recovery from attack is essential; need detection/isolation/fallback in spite of persistent threat RJ: network segmentation would help Rudy: Need resilience/recovery; can’t get completely hardened RJ: software updates are critical; need secure OTA updates Rudy: need ECUs with TEE capabilities and secure (unique) key distribution Rudy: safe-mode: limp home mode, slow with hazards flashing lights RJ: this happens with sensor malfunction Rudy: limp home mode differs between manufacturers; maybe 2nd gear; due to engine sensor problem, transmission failure RJ: A/C shuts off; could engage via cyber attack RJ: Cummings engine ECU does checksum at startup RJ: modern vehicles don’t “turn off completely” ; phone app can communicate with vehicle even when you’ve turned it “off” Rudy: telematics unit stays up for a few hours; turns off after a few days; SMS is first service available when modem connects to base station Rudy: wakes up every few seconds (programmable); Qualcomm chip sets
[video]
[discussion of Sensoris video]
Cheryl: those who opt in need to make their data available and cannot hoard. consent and neutral server concepts in discussion
Jennie: in 2016 when we started
we had 40ish loads per day
... I want to introduce freight specific terminology
[Load]
Jennie: they are essentially same
as Uber rides
... customers for Uber rides is individuals, for freight it is
companies
... we partner with carriers to fullfil shipments
... we knew we wanted first class freight models and handle how
they differ from Uber rides
[early data model]
[assumptions]
[carrier]
Jennie: only 5 at first, now up
to 63 fields
...started in 2016 as
MVP, then launched in 2017
...This is diffferent
from uer ride, but similar. Longer duration, higher price and
customers are businesses
...Uber Freight partners
with carriers to fullfill shipments
...First model was a
simple ; they assumed 1 load = 1 carrier & 1 load = 1
source + 1 destination
Jennie: we experienced pain points pretty quickly, needed a load to be handled by multiple carriers etc
[fastest growing logistics company in US]
Jennie: you can imagine the use
cases we need to support today are more complex than in the
past
... how do we refactor and grow along with the business
... Jon has some new use cases to share
Jon: with all this growth comes challenges, new opportunities and strategies
[load]
Jon: load and job model. load has
5 main abstractions, commodities being shipped. items are
grouped to a purchase order
...First model was a
simple ; they assumed 1 load = 1 carrier & 1 load = 1
source + 1 destination
...This had limitations;
there was no distinction between shipment obligation and
fulfillment strategy
...no support for:
shipments with multiple stops; shipments with actions at the
end of trips (either required by uber or the shipper)
Jon: job, tasks
[new model]
[multi-stop]
Jon: seeing some major gains. we
have PO as a first class citizen
...the fastest growing
logistics company in US history
...The model now has two
large domains : Load Model and Job Model
...Load Model has
Purchase Order // Item // Stop // Stop Task
...Job Model has Job //
Waypoint // Waypoint Task
...the new model has
already decoupled shipment obligation (load) and fullfillment
(job) and now can have multiple pickup or dropoff operations at
a single place
Jon: we can now handle more than two stops per shipment
[breakdowns]
Jon: original job is driver
arriving at point 1, transporting it to 2 destination
... we wanted a really good audit log of how this happened
[Truck order not used]
Jon: driver arrives but shipment
missing or not ready
... a single carrier may have multiple drivers for a
shipment
[Bundles or chains]
Jon: more complex models where we
group jobs, optimize routes
... may be priced differently than single shipment jobs
[relays]
Jon: multiple carriers and possibly an exchange, differs from multi-driver in trailer is switched or unloaded and repacked into next container
[powerloop]
Jon: this is more about moving trailers around, empty or full for optimization
[cross border]
Jon: carriers may not be certified to cross borders which may necessitate a change in carrier
[expedited]
Jon: some customers may want
things delivered quicker and beyond the parameters of ERODS
(driver hours)
... (HOS)
[intermodal]
Jon: we want to include multiple
modes for a shipment, could include rail, ship, truck etc
...This enables some new
use cases:
...multi-stop
...breakdowns
...bounces (need to
remove a carrier from responsibility of shipping)
...Truck Order Not Used
(shipment is ready but when driver arrived there is no shipment
to pickup)
...multi-driver (single
carrier wants to use two drivers)
...bundles/chains (a
grouping of jobs that can be advertised as a single unit --
gives uber the power of advertising multiple jobs as a
unit)
...relays (mulitple
carriers execute one shipment, performing an exchange in the
middle)
...trailer swap (two
drivers pass each other and swap trailers -- keeps drivers
local to a region)
...PowerLoop (uber has a
partnership with another entity whose goal is to move items in
cycles to create more fluid markets of goods -- enables uber to
pickup and drooff empty trailers)
Jon: those were the primary use
cases we wanted to tackle
...Cross-border (not
every carrier has the permission to cross the national
borders)
...Expedited (team
driving -- modeled as a two driver exchange)
...Intermodal (e.g. boat
to truck)
George: you handling different sized loads, refrigerated, etc
Jon: we have load requirements that do take that into consideration
Joseph: this dynamic or fixed
pricing?
...The majority of their
shipments managed by Uber are full truck load
Jon: there are some dynamic aspects in the pricing
Jay: some loads treated differently, more challenging and priced differently?
Jon: we are not getting into hazardous loads, but categorizing more difficult ones
Jason: what is the tracking, progress etc?
Jon: a single load can have multiple actions
Luis: what kind of db are you using for this?
Jon: relational for now and may choose something else later, primarily transactional
[hotshot trucking]
Jay: how did you decide on your initial set of use cases?
Jon: based on having the right
people in the room, learning from those with freight
backgrounds
... what are the critical use cases for our business goals.
expanding into a new freight modality was more important than a
new geographic market
... we wanted to go deep on a strategy to learn from it
George: this using h3 spatial model for load balancing
Jon: our marketplace is pretty much from scratch
Ted: purchase order terminology seems off, sounds more like shipping manifest. some of these are variations and hope you first commonalize and then deem customization to avoid duplicating definitions
Jon: yeah, we will be going that
way
...we could recomend to
w3c that a new group should be formed on coordinating
automotive data standardization (then lists all the steps -- it
will take a while)
...an idea is leverage
the fact that the geospatial group is up for re-chartering: use
it as the platform for making the coordinating auto data
standards
...someone should
catalog these other ontologies. schema.org is a good starting
point.
...There were a small
number of position statmeents -- but many projects and
ontologies shared.
...We need to go through
all the link dumps and identify where we can unify
...If we don't get the
ontologies setup early then there wil be lots of work to do in
the future on interop and that work will be effort that could
have been saved.
Ted: @@outline
George: you identified area of
common interests
... we got together and talked, are there things we want to
work on
... can we get these organizations to work together?
... as for the entities list, yeah there are things I want to
work on
Josh: I think the breakout groups
were all in scope
... accessibility track from yesterday too
...george asks what is
the scope of what we will work on? a: the items circled on the
whiteboard from the standardized entities for transport are
topics that he wants to work on
George: formal coordination takes
time
... the more parties you try to coordinate the more difficult,
sure aspire to that but meanwhile people should continue on
current efforts
... similarly to get OGC engagement, need to see if there is
interest
...george suggests to
take each of those entiteis and identify who has a pieve of
each and then compare who has what and try to harmonize
Peter: part of the Auto WG and need to see the ontology being worked on get uptake
Mark: from JTC-1, there are a
number of entities on the board that should be a direct
focus
... share the view we do not want to wait on established
coordination
...george suggests to
take each of those entiteis and identify who has a pieve of
each and then compare who has what and try to harmonize
...volvo says that they
are working on VSS right now and they really need an ontology
on top of that _yesterday_
...mark shares the view
that waiting on formal coordintion isn't possible. this needs
to be worked on now
Carlos: we want to get moving and
need it moving now too
... we have a number used internally and agree to the list on
the board, see what we want to work on
Ken: first and foremost we need
to identify major entities that need to agree on, two
organizations either started on the area or should take lead
and that will help us move quickly
...ken asserts that
first and foremost we need to identify the major entities on
which we need to agree. then identify people to lead those
entities
Mark: Ken and I will put together
a document
...if we want this all
to come together then each group needs to use the same modeling
rules
...setting baseline
concepts early on will make it more likely to have all the
models come together
...Mark and Ken will be
drafting a document on modeling rules
...Mark asks if we can
use the email list. Ted will create a mailing list.
...Ted has another idea:
pick on entity that is a top priority item. Start looking at
defining what are the criteria of teh work that needs to be
setup. e.g. who is the lead? How do we influence it?
...Ted suggests Route ;
Josh says 'trip' for Uber byt Route is very similar
George: difference between trip and route?
Josh: trip is a route taken including timestamps
Ken: trip could be multi-modal
trip, route, observations...
George: routing api would be a really good thing
Josh: we are doing quite a bit with apis already
David: @@737
... as we increase abstraction remember the impact to the
physical world
@@mailing list, report, assessment of route models suggested as an early starting point
Jason: @@
* Jason @ here suggests that the level of verbosity and complexity should vary between the application of the data model. e.g. VSS is simple a low-verbosity and fits in the embedded space; but at loud exchange the format might need to be more verbose and ready for higher throughput