Quarterly Report for MassDOT & City of Boston 2nd Quarter 2021

Background

Motional is committed to designing for people: operation in Las Vegas, which has safely for families that need to get their children to provided more than 100,000 self-driving rides school safely; for elderly passengers who need to members of the public. We are proud to continued access to mobility; and for urbanites report that our attention to safety has extended who, more than ever, have a choice in how they into our real-world operations. We have driven get around cities. We know that self-driving over 1,000,000 miles in complex city vehicles have the potential to bring vast environments worldwide while maintaining a benefits to humanity: increased mobility, fewer record of zero at-fault incidents. traffic-related deaths, and a greener planet. But the only way to fulfill these promises of Today, our global team—spanning the U.S. and tomorrow is to build trust in the technology Asia—is dedicated to delivering safe and today. We believe that when we demonstrate reliable production-ready SAE Level 4 robotaxis openness and collaboration, trust follows. that will make roads safer and improve mobility worldwide. As we advance the technology, our Our team’s expertise in autonomous driving people-first ethos will ensure that safety, can be traced from our R&D roots at MIT and security, and privacy are embedded in every Carnegie Mellon University, where we step. showcased our autonomous technology in the DARPA Grand Challenge and DARPA Urban Motional has made the most of the first year Challenge, to our present-day commercial since it was established as a joint venture 1

between Hyundai Motor Group and Aptiv in March 2020. We’ve made significant strides in Operational Design Domain (ODD) advancing our driverless product. Motional has concluded over 100,000 hours of safety Our vehicles are designed to operate in evaluations, built processes for actively low-speed (<35 MPH), urban environments in identifying cybersecurity threats, and pursued various conditions. We continuously validate all inclusive User Experience (UX) research to vehicle performance and behavior changes to ensure our vehicles will ultimately be our AVs in simulation, then in a closed-course accessible to disabled riders. After a rigorous setting before operating them on public roads. review by a third party assessor, we began To date, we have experience testing on public testing our Pacifica in driverless mode on the streets with a variety of road actors, including public roads around Las Vegas in early 2021. In heavy vehicle traffic, cyclists, and pedestrians. March 2021, we announced that we’ll be Additionally, we have operated our AVs safely migrating to an all-electric Hyundai platform, in daytime and nighttime and windy, rainy, and the IONIQ 5, for our AV testing. We also have snowy conditions in closed-course and public launched our nuPlan benchmark to provide a road environments. dataset for autonomous vehicle planners, while working to safely resume passenger service in Amount of testing Las Vegas in line with the easing of COVID-19 restrictions. Our testing occurs primarily during regular business hours (Monday through Friday, Testing activity 9AM-5PM). As mentioned above, this testing includes specialized testing in closed-course and data gathering in the Seaport / South Motional has started to broaden the scope of Boston area. its autonomous capabilities by continuing to work on suburban elements as well as taking As we continue to develop and build out our aim on more urban dense driving. For example, Generation 2 Ioniq platform we are exploring by utilizing our closed course tracks we are multiple avenues for more efficient testing working on traffic light environments and methods. Such methods include finding better ensuring consistent perception and reaction to ways to verify the basic autonomous changing traffic lights. capabilities of the vehicle before doing significant real world testing. These new Our work on the Safety Steward role is practices involve multiple stakeholders within continuing to grow and develop. This individual Motional to ensure that our approach is not will sit in the front passenger seat to oversee only safe but also allowing ourselves to be driverless testing. We are refining the protocols more time effective. for Safety Stewards, including when and how they intervene with the emergency stop buttons. 2

Takeover procedure cases in which the AV would have handled the situation without incident. Vehicle Operators take over manual control in any situation in which they feel uncomfortable Description of ADS System failures or unsafe. Planned takeovers are also done when finishing a mission or approaching We did not experience any unanticipated situations that are not within the outlined ODD. failures or disruptions while driving in We are also refreshing our Fault Injection autonomous mode. As we explain above in training with all Vehicle Operators where greater detail, in specific traffic scenarios, our intentional system errors are introduced to vehicle operators take over manual control make sure our operators takeover in the proper because of known limitations of the current fashion before returning to public roads. state of AV software.

During the Second Quarter, our vehicle Goals for future testing operators took over manual control of our AVs

in the following situations: Continue to expand our autonomous capabilities through proven closed course track ● When emergency vehicles were in active tests before transitioning to public road driving.

operation (e.g., sirens and lights We anticipate being on public roads in both our

activated) in the roadway; Pacifica and Ioniq platforms this year. ● When law enforcement officers were manually directing traffic in When looking at future testing we are keeping intersections through which our AVs an eye on ride-hail abilities and much of that were traveling; comes down to pick up and drop off scenarios. ● When construction vehicles were While our Product team is exploring the various obstructing our lane of travel; interactions members of the public would have ● When oncoming vehicles or bicycles with a fully driverless car in their own User violated lane boundaries; Experience testing, the Operations team will be ● When weather conditions deteriorated working on the physical practice of rapidly; and, autonomously doing pullovers in challenging ● When other vehicles were exhibiting scenarios that we see in the Boston area. erratic behavior near our AVs. Insights A vehicle operator’s decision to take over manual control in a given situation does not Motional’s autonomous vehicles make necessarily indicate that continued

autonomous operation in those situations rule-based decisions derived from the vast would be unsafe. Because we instruct our streams of data that they receive from vehicle operators to err on the side of caution, advanced LiDAR remote sensing technology, we expect that takeovers will occur in many cameras and radar. Terabytes of data, some of it extraneous, some of it critical to safe driving, 3

are broken down by our software using deep also vulnerable road users and physical learning. Data from all of our sites is constantly objects. The pilot program starts with offline broken down and analyzed in an ongoing comparisons of the perspectives from the two iterative process that directly informs our sets of cameras and was designed to transition on-the-road testing in Boston. Our vehicles are to real-time transmission of data from the Derq thus not encountering the world anew each computers to the vehicles. This technology is time they travel from the testing lot along our especially promising for extremely complex routes throughout the Seaport area. The intersections where even human drivers production of detailed, high-definition mapping struggle to predict the activity of massed is a key part of our development process, as vehicles going in multiple directions. In the we develop and annotate maps which provide long term, this kind of technology could reduce a firm foundation for our vehicles as they accidents at problem intersections like navigate through their Operational Design Columbia Road and the Southeast Expressway Domains (ODDs). Our publicly released in Dorchester. nuScenes dataset, which has been cited more than 220 times and includes many annotated V2X technology is an important pillar of the images from our mapping of Boston, gives future integration of autonomous vehicles with insight into how we interpret data from a connected environment that includes the Massachusetts. Internet of Things. Given the recent federal reallocation of a significant fraction of the Motional’s testing is getting smarter, as we trial mid-band DSRC (Dedicated Short Range new ways of connecting our vehicles to even Communications) spectrum previously richer sources of real-time data. Our reserved for connected vehicles to unrestricted relationship with the predictive analytics commercial use, autonomous vehicle platform Derq has facilitated a companies have had to move quickly to turn to vehicle-to-everything (V2X) pilot in Las Vegas V2X and long-term evolution (LTE). Innovations with important implications for the future like the Las Vegas roadside camera pilot development of connected and autonomous program speak to the power of V2X to enable vehicles. Cameras located atop roadside smarter driving by autonomous vehicles intersections feed data to Derq’s roadside maneuvering in complex intersections. computers, which can communicate an entirely Motional is closely tracking a number of new view (a birds-eye view) to Motional’s ongoing V2X studies underway in the Boston vehicles that complements the onboard area and the Commonwealth as a whole. sensors. Real-time analysis of intersections gives an unobstructed view of traffic to the Eventually, V2X can not only support safe autonomous vehicles, including information driving, but also lead to more smooth and not just about other cars in the intersection but comfortable rides for passengers by enabling 4

the vehicle to make earlier decisions. local testing operations, hiring of fewer vehicle Connected vehicles could be alerted to the operators and reduced public exposure to presence of objects out of their direct line of innovative AV technology. sight. Fleet and traffic management will both benefit from this technology, which will enable Other consequences of overreliance on operators and urban planners to make smart, imprecise and non-uniform disengagement real-time decisions about where to put reporting include the generation of datasets resources. While the pilot program in Las which produce research that then informs Vegas is an incremental step towards the public policy. For example, a study on vehicle connected infrastructure of the future, it operator reaction times during speaks to the disruptive potential of V2X disengagements in California was produced at technology in Boston and globally. a time when seven companies were conducting driverless testing in the state. But Feedback for municipal and because only two companies reported precise reaction times for disengagements, while the state transportation others chose to report reaction times in terms engineers, planners, and of upper bounds (e.g., <1 second), the study’s reaction time dataset was limited to those two policymakers operators. However, the industry as a whole is affected by the policy recommendations that Disengagement reports from autonomous are drawn from this kind of study, even though vehicle companies are attractive metrics for the most meaningful data speaks only to the state and local policymakers given the best practices and state of the technology at seemingly clear insight they provide into the two companies in a field where resources, reliability of driverless systems. The strategy, and technical proficiency diverge autonomous vehicle industry has raised a widely. number of issues with disengagement reports, including that the degree of discretion afforded In considering whether and how to utilize to operators in defining disengagements disengagement metrics, localities should means that true company-to-company consider the precision with which comparisons are impossible. First-order disengagements are defined as well as how consequences of overreliance on even incomplete disengagement data will be disengagement reports in a locality can include publicized and used by the media, the business an incentive for operators to conduct less community, academic researchers, and public road testing volume overall as well as ultimately policymakers. less difficult testing, as well as more testing on private tracks. The result is less investment in