Automation presents the potential to save many lives on our streets and highways, but it also presents many challenges. This presentation is about two aspects of those challenges – those that can benefit from decades of automation experience in aviation, and those for which there is little experience because they have not generally been encountered before.
Aviation. More specifically, the presentation describes aviation experience that has demonstrated the importance of “Human-centric” automation, as opposed to automation “because we can.” Aviation has also shown that complete automation, i.e., with no human present, will not be capable of safe operation until automation designers develop “graceful exits” if (a) the automation fails or (b) the automation encounters unanticipated circumstances. Finally, as aviation automation has become more reliable, it has shown that humans are not good monitors of reliable systems.
Other. The presentation also describes several aspects of automation that have not previously been encountered or addressed in aviation. Included among that list are the importance of street testing; the lack of training for drivers; the frequency of software updates; the use of software that learns with experience; the relative ease of cyber attacks; the lack of federal standards; the competition between automakers re safety; and the need to address ethical concerns.
Over the past decade, our research has focused on algorithms that enable increased vehicle autonomy including vision, control and learning algorithms. We demonstrated these algorithms on different vehicle types ranging from drones and off-road driving vehicles to mobile manipulators. More recently, our team was selected as one of eight teams to compete in the North-American SAE AutoDrive Challenge, a three-year self-driving competition sponsored by General Motors among others. For this competition, we built a self-driving car in six months, took it to the Year 1 competition in Arizona in April 2018, and won!
The automation of vehicle behavior is particularly challenging when the vehicle operates in increasingly unpredictable and changing environments. Traditional robot algorithms largely rely on a-priori knowledge of the system and the environment, which is insufficient when requiring vehicles to deal with unseen situations. This is also reflected in self-driving car algorithms and explains why highway driving is much easier than city driving.
This talk aims to highlight current advances and limitations of self-driving technology with the goal of bridging the gap between user interface design and the current capabilities of self-driving technology. As self-driving technology is evolving rapidly, I would also expect automotive interfaces to be required to evolve quickly.