Fair Play in the Fast Lane: Integrating Sportsmanship into Autonomous Racing Systems

TL;DR

The paper addresses the gap of enforcing sportsmanship in autonomous versus racing by proposing a bi-level game framework that couples a high-level Stackelberg-based intention planning with a low-level Generalized Nash Equilibrium trajectory planning. The approach uses Monte Carlo Tree Search to derive intention-level strategies and a GNEP with an Iterative Best Response solver to produce SPS-compliant trajectories, ensuring blocking and overtaking adhere to predefined rules. Results from straightaway and corner scenarios show that explicit SPS constraints can balance competition and safety, reducing unsportsmanlike behavior while preserving performance. This work provides a formal, extensible foundation for ethical autonomous racing and suggests avenues to scale with deep learning for more complex multi-agent environments.

Abstract

Autonomous racing has gained significant attention as a platform for high-speed decision-making and motion control. While existing methods primarily focus on trajectory planning and overtaking strategies, the role of sportsmanship in ensuring fair competition remains largely unexplored. In human racing, rules such as the one-motion rule and the enough-space rule prevent dangerous and unsportsmanlike behavior. However, autonomous racing systems often lack mechanisms to enforce these principles, potentially leading to unsafe maneuvers. This paper introduces a bi-level game-theoretic framework to integrate sportsmanship (SPS) into versus racing. At the high level, we model racing intentions using a Stackelberg game, where Monte Carlo Tree Search (MCTS) is employed to derive optimal strategies. At the low level, vehicle interactions are formulated as a Generalized Nash Equilibrium Problem (GNEP), ensuring that all agents follow sportsmanship constraints while optimizing their trajectories. Simulation results demonstrate the effectiveness of the proposed approach in enforcing sportsmanship rules while maintaining competitive performance. We analyze different scenarios where attackers and defenders adhere to or disregard sportsmanship rules and show how knowledge of these constraints influences strategic decision-making. This work highlights the importance of balancing competition and fairness in autonomous racing and provides a foundation for developing ethical and safe AI-driven racing systems.

Figures (6)

• Figure 1: Sportsmanship in autonomous racing. Left: Under the one-motion rule, the defender is allowed to make only a single lateral move to block the attacker. Continuous swerving to defend its position is prohibited. Right: Under the enough-space rule, when the attacker attempts to overtake along the edge of the track boundary, the defender must leave sufficient space. Deliberately cutting in to force the attacker to brake or risk a collision is not permitted.
• Figure 2: Visual representation and pertinent parameters in a car racing scenario.
• Figure 3: Pipeline of the proposed algorithm. In the high-level game, the MCTS algorithm is applied to search for optimal strategies with respect to racing intention. In the low-level game, GNE trajectories are obtained conditioned on high-level intentions, which are obtained through forward propagation of high-level MCTS. Utilities are then determined and provided back to the high-level game solver for back-propagation.
• Figure 4: Results on the straightaway under SPS1. In (a) where none of the players know the existence of SPS1, the defender (racing car in red) switches to the front of the defender (racing car in blue) to perform a blocking. Since the attacker does not know the existence of SPS1, it considers overtaking to be impossible and thus decides to follow the attacker. In (b) where both players know SPS1, the defender also performs blocking to delay the overtaking, while the attacker decides to switch to the lower half of the track. Due to SPS1, the defender knows that a second blocking is forbidden, so it stays in the upper half of the track to let the attacker pass. In (c) where only the attacker knows the existence of SPS1, it attempts to perform an overtaking by switching to the lower half of the track after the blocking, similar to the case in (b). However, since the defender does not know SPS1, it continues to block the attacker, and thus the attacker fails to overtake. In (d), although the defender knows SPS1, the attacker does not, so it cannot come up with the correct strategy as in (b). Therefore, it also fails to perform an overtaking.
• Figure 5: Simulation results at the corner under SPS1. The results are similar to the straightaway case, such that the attacker manages to perform the overtaking if and only if both players know the existence of SPS1 (as is shown in (b)). Note that results in (c) are slightly different from that in the straightaway case. After the defender switches to the inner lane for blocking, the attacker switches to the outer lane for an overtaking attempt. Since a longer distance is covered along the outer lane, the intended overtaking is delayed until both players exit the corner. Therefore, the best strategy of the defender in response is to stay in the inner lane and perform a second blocking only after both players exit the corner (compared to an immediate second blocking as in Fig. \ref{['fig:strateghtSPS1']}.(c)). This strategy will grant the defender longer progress along the track.

Theorems & Definitions (4)

• Definition 1
• Definition 2
• Definition 3
• Definition 4