AI, Cognitive Vehicles, Interviews, Smart Mobility

Beyond Perception - Interview with Dr. David Atkinson, Continental

Dr. David J. Atkinson
Head of Systems & Technology and Chief Research Scientist for Artificial Intelligence, Continental AG

Dr. David J. Atkinson
Head of Systems & Technology and Chief Research Scientist for Artificial Intelligence, Continental AG

About David:

Dr. David J. Atkinson is Head of Systems & Technology and Chief Research Scientist for Artificial Intelligence at Continental’s Silicon Valley Research & Development Center. He has held this position since May 2017 and oversees systems and technology projects for future transportation and mobility systems, with a focus on intelligent driver assistance systems and autonomous vehicles.

Atkinson’s career spans basic research in Artificial Intelligence to senior executive positions in management of research and technology. Before joining Continental, he worked at the Florida Institute for Human and Machine Cognition, where he led research projects in the area of autonomous robotic systems. Previously, Atkinson worked for NASA’s Jet Propulsion Laboratory at California Institute of Technology in multiple positions of increasing technical management responsibility. He was a founder of NASA’s Artificial Intelligence program and made significant contributions to spacecraft autonomy, planetary exploration robotics and spacecraft control center automation. An entrepreneur in the early 1990s, Atkinson was co-founder of a successful startup that provided Internet-based supply-chain management services to the electronics industry.

The Pop in your Job: What drives you - What is your passion?

The application of artificial intelligence for intelligent assisted and autonomous vehicles is very motivating for me because it represents an opportunity to save lives and bring enormous social benefits to people all over the world. Moreover, it creates an environment for new ideas and that is the greatest thrill of all.

we.CONECT: What are your main responsibilities in your current role?

Dr. David J. Atkinson: I am responsible for systems and technology projects for future transportation and mobility systems in the Silicon Valley R&D Center,with a focus on intelligent driver assistance, autonomous vehicles, and smart city traffic infrastructure. I also track technology trends and emerging mobility companies, and serve as their engineering contact with Conti. I am the Chief Scientist for Artificial Intelligence in the Chassis & Safety Division of Continental and I work closely with our worldwide corporate efforts in this technology area.

we.CONECT: What fascinates you most about autonomous driving?

Dr. David J. Atkinson: The technical challenges of course, and the challenge of pulling together all the diverse technologies, stakeholders, and more in an exciting commercial sector. I have been working on autonomous robotics since 1984, at first for NASA, later for other government agencies. I’ve performed my own basic research, led teams, fielded successful applications of AI and managed large international research programs in AI and robotics. So, I know the technology, including what most people don’t see yet because it is still emerging in university research labs. I am fascinated that all the pieces are finally coming together to create intelligent, autonomous robots that can help solve critical social problems, such as safe mobility.

we.CONECT: What are your predictions for autonomous driving?

Dr. David J. Atkinson: Autonomous mobility will be at least as disruptive and as valuable an innovation for society as the smartphone.

we.CONECT: What role does AI and cognitive computing play in self-driving car technologies?

Dr. David J. Atkinson: Without AI and associated technologies, self-driving cars are not possible. AI is integral to the vehicle’s ability to sense the world, to perceive objects, understand other actors (vehicles, people), the environment and to reason. AI is essential for reacting appropriately and rapidly to dynamic events during driving. The vehicle’s AI deliberates on the task of driving, creates plans, and executes behaviorsthat achieve driving tasks. The complexity and uncertainty of the real world means that vehicle autonomy technologies must bootstrap themselves (with our help) to a basic level of driving competence using machine learning. This is true primarily in the area of sensing, perception and reflexive (low-level) behavior such as lane following and obstacle avoidance. Eventually, I expect machine learning to continue over the operational lifetime of a vehicle. It will learn how to be a better driver, and learn about the world, about people, and about itself.

we.CONECT: There is a lot of confusion around autonomous driving. What are the different routes to level 5 and what are the biggest challenges to get there in the short term (in the next 5 years)?

Dr. David J. Atkinson: Level 5 is hard, but not confusing. We’ve already achieved that level of autonomy in spacecraft, aircraft, off-road vehicles, undersea vehicles and other robotic systems. Autonomous driving on roads is hard because there is not much space to move, situations and environmental factors are uncertain, and there are many other actors who may behave unpredictably. These challenges are all surmountable. Level 5 is an engineering challenge, not a scientific challenge.

Level 4 is confusing primarily because we have yet to settle on the respective responsibilities of human driver and vehicle. It is difficult to say what humans or vehicles MUST do or MAY do without reference to specific operations scenarios, and there are a near infinite number of particular scenarios. The big challenge is shared, cooperative control authority.

we.CONECT: How is your company developing deep learning capabilities? What are the challenges?

Dr. David J. Atkinson: We have dedicated groups at the corporate level who are out in front. They are applying multiple machine learning techniques to various problems of special interest to Continental. They work in concert with technologists such as myself in the Divisions and we all work to support technologists and product developers in our business units, ADAS in particular with respect to automated driving.

The engineering of machine learning capability is still, in part, an art and not a well-defined process. It requires expertise created from hard won successes, not just coursework. There are not many experts around, so we are both hiring and growing our own experts. One major challenge is predictability of ML as an engineering process. Machine learning is hard, and first attempts are generally not good enough. We need to build a record of success so that we are confident in repeatable, high quality machine learning processes that yield valuable results. Another major challenge is test, verification and validation. ML and AI disrupt traditional engineering approaches. We are working actively on this.

we.CONECT: Please explain in brief the key aspects of your session at the Auto.AI USA 2018.

Dr. David J. Atkinson: The headline is “Beyond Perception: AI Reasoning for Autonomous Vehicles”. I intend to be provocative. Level 4/5 autonomous driving requires driving capability that reliably handles the most challenging driving scenarios. Vehicle functions must include automatic reasoning about alternative choices of vehicle behavior, and decisions made under time pressure and conditions of uncertainty. I will discuss a few requirements, relevant technologies, and other design considerations.

we.CONECT: What expectations do you have towards the Auto.AI 2018?

Dr. David J. Atkinson: I hope for tough questions that will perplex me. I expect that I will learn quite a bit from the other attendees and speakers.

we.CONECT: Thank you very much for your time to participate in an interview.