Conference Presentation, Keynote
16 Questions About Self Driving Cars
- The speaker projects that the market for autonomous mobility will expand significantly due to collapsing costs, with cars eventually generating higher revenue than phones despite fewer total units, while non-automotive items like planes, trucks, and social robots (e.g., Jackrabbit) will also move toward autonomy.
- Regarding timelines, a personal expectation of the self-driving car world being fully realized exists within two years, though industry-specific forecasts vary: Google aims to launch in Singapore by 2018 and reach top 10 cities by 2020; Delphi and Mobileye project system availability by 2019; GM and BMW target 2020 and 2021 respectively for autonomous vehicles and the iNext model; Ford anticipates Level 5 fleet availability by 2021; Tesla predicts readiness by 2023; Uber expects a fully autonomous fleet by 2030; and the IEEE projects 40% of cars on the road will be autonomous by 2040.
- Incumbent auto manufacturers are expected to pursue an incremental strategy of adding features one at a time rather than immediately adopting Level 5, whereas Silicon Valley competitors like Google, Tesla, and Ford are predicted to potentially launch at Level 4 or 5 to avoid mixed-mode user experience failures.
- Technology costs are forecast to drop dramatically, with LIDAR transitioning from $75,000 to $250 as solid-state technology replaces bulky form factors, potentially allowing self-driving capabilities to be added to a $17,000 car for an additional $2,500, while social robots are expected to reach a $500 price point in 5 to 7 years.
- Technical approaches will likely involve a blend of deep learning and traditional control systems, utilizing data from real-world driving and virtual reality simulations, with an estimated need for 10 million miles of driving data for neural net convergence, though companies like Tesla are attempting systems without LiDAR.
- Sensor and mapping strategies will address the need for high-definition maps, which may be limited by a global bottleneck of only three mapping companies, leading to a potential shift toward real-time "on the fly" calculation requiring increased power envelopes ranging from 50 to 500 watts.
- Safety metrics anticipate an initial short-term increase in accident rates due to mixed human and autonomous driving, but long-term full autonomy is expected to reduce accidents to near zero, with the speaker citing that 24 out of 25 fatal accidents are currently caused by human error.
- Infrastructure evolution includes the implementation of V2X communication, with Mercedes-Benz integrating V2V radios in the S series starting in 2019, which aims to eliminate the need for traffic lights by allowing cars to negotiate intersections like data packets, though security concerns like spoofing and deployment delays remain.
- Market dynamics predict a shift in consumer loyalty from car manufacturers to fleet providers like Lyft and Uber, transforming automakers into B2B entities similar to the airline industry, while demand curves post-adoption and second-order social effects remain difficult to predict.
- Regulatory and economic shifts include the possibility of making human driving illegal if algorithms are proven superior, insurance premiums becoming a function of algorithm effectiveness rather than demographics, and complex liability questions arising from hacking scenarios involving garage doors and homeowners.
- Future vehicle design for Level 5 autonomy will feature no steering wheel and no manual control capability, while commute times may fluctuate due to induced demand or savings from reduced parking and repair shop requirements.
- Specific company strategies note Chinese manufacturers aggressively pursuing the market due to high volumes of deep learning papers, while Google is viewed as having transparent reporting on accidents, and the speaker identifies Tesla as the "obvious bet" among Silicon Valley firms to dominate the space.