Learning overtaking and blocking skills in simulated car racing

Abstract

In this paper we describe the analysis of using Qlearning to acquire overtaking and blocking skills in simulated car racing games. Overtaking and blocking are more complicated racing skills compared to driving alone, and past work on this topic has only touched overtaking in very limited scenarios. Our work demonstrates that a driving AI agent can learn overtaking and blocking skills via machine learning, and the acquired skills are applicable when facing different opponent types and track characteristics, even on actual built-in tracks in TORCS.

A Survey of Planning and Learning in Games

“…DRL, in particular, has become quite successful in this domain, since the introduction of DQN [16], a well-known algorithm that managed to achieve human level gameplay and above on several games from the Atari 2600 gaming platform, using raw images as input. Examples of this include work in areas as diverse as deriving adaptive AI agents in RTS games [48], developing natural behaviors for companion NPCs [50], blocking and overtaking in car racing games [100], learning from demonstration [101][102][103], simultaneous learning (i.e., human feedback is introduced during the RL process) [104,105], simulating crowds [8], procedural content generation [106], and solving the micromanagement task in RTS games [107,108].…”