Cognitively-inspired episodic imagination for self-driving vehicles
Published at IROS 2019, Workshop: “Towards Cognitive Vehicles: perception, learning and decision making under real-world constraints. Is bio-inspiration helpful?”
The controller of an autonomous vehicle needs the ability to learn how to act in different driving scenarios that it may face. A signiﬁcant challenge is that it is difﬁcult, dangerous, or even impossible to experience and explore various actions in situations that might be encountered in the real world. Autonomous vehicle control would therefore beneﬁt from a mechanism that allows the safe exploration of action possibilities and their consequences, as well as the ability to learn from experience thus gained to improve driving skills. In this paper we demonstrate a methodology that allows a learning agent to create simulations of possible situations. These simulations can be chained together in a sequence that allows the progressive improvement of the agent’s performance such that the agent is able to appropriately deal with novel situations at the end of training. This methodology takes inspiration from the human ability to imagine hypothetical situations using episodic simulation; we therefore refer to this methodology as episodic imagination. An interesting question in this respect is what effect the structuring of such a sequence of episodic imaginations has on performance. Here, we compare a random process to a structured one and initial results indicate that a structured sequence outperforms a random one.