PhD position Risk-aware motion planning

While automation offers opportunities to make society safer, it comes with new risks, some of which are fundamental and, others more technological. Autonomous agents require an understanding of how humans respond to the uncertainty and risk that they bring. From the agent’s perspective, actions are taken to bring itself into safety if there are uncertainties, but this may introduce new risks for others. For example, a vehicle that comes to a stand- still at a strange location, or reduces speed for a green traffic light, has already resulted in accidents, simply because this does not match human expectations. It is a major challenge to develop agents and frameworks that account for uncertainty, risk and interaction in the way humans do.

As human behavior depends on a tightly controlled perception-action cycle that carefully considers uncertainty and risk, so should the behavior of an autonomous agent. But this is not trivial to attain. For example, the dynamics of the environment in which it operates can be unpredictable, and control options may be unavailable or have limited ability. Inspired by how human brains deal with uncertainty, in this project you will develop probabilistic frameworks for motion planning in autonomous agents, such as cars or teams of drones. We will work on a fundamental understanding of how autonomous agents can cope with uncertainty and provide means for computing performance guarantees of autonomous AI agents under uncertainty, which will be integrated to various degrees into a use-case with self-driving shuttles. In autonomous agents, this probabilistic framework consists of three successive stages: processing sensory information to arrive at state-estimation (what is the agent perceiving and where), determining action-plans from this state estimation (what the agent can and should do, and how much risk it should take), and converting abstract action-plans into precise motion-plans. In your PhD, you will focus on the second challenge, namely the generation of safe risk-aware motion for the autonomous robot. You will collaborate with a second PhD candidate focused on estimation.

You will be expected to test your algorithms in practice with mobile robots (ground or aerial) and a self-driving car, in collaboration with 2getthere.

For your PhD you will be embedded within the Autonomous Multi-Robots Lab in the Department of Cognitive Robotics at TU Delft.

The main focus of the Cognitive Robotics department is the development of intelligent robots and vehicles that will advance mobility, productivity and quality of life. Our mission is to bring robotic solutions to human-inhabited environments, focusing on research in the areas of machine perception, motion planning and control, machine learning, automatic control and physical interaction of intelligent machines with humans. We combine fundamental research with work on physical demonstrators in areas such as self-driving vehicles, collaborative industrial robots, mobile manipulators and haptic interfaces. Strong collaborations exist with cross-faculty institutes TU Delft Robotics Institute and TU Delft Transport Institute, our national robotic ecosystem (RoboValley, Holland Robotics) and international industry and academia. http://www.cor.tudelft.nl/

For information about the Autonomous Multi-Robots Laboratory at the Delft University of Technology, see https://www.autonomousrobots.nl/ .