Beyond Self-Play and Scale: A Behavior Benchmark for Generalization in Autonomous Driving

Figure 2 : Qualitative comparison of our Interactive1k and Random1k splits. Four representative scenarios from each split, showing the ego vehicle (red), surrounding agents (blue), the ego goal (green), and the expert trajectory (red). Interactive1k (top row) is dominated by dense urban situations while Random1k (bottom row) reflects simpler cases.

Figure 5 : At-fault collision rate ( \% ) on the Interactive1k split for PPO planners trained via self-play under a simple (top) and complex (bottom) reward formulation, evaluated at intermediate checkpoints up to 10^{11} agent steps against PPO (solid) and IDM (dashed) traffic. Both reward formulations yield low collision rates against their own behavior but fail to generalize to IDM behavior.

Figure 12 : Qualitative comparison of traffic agent models. We visualize rollouts of IDM, PPO, SMART, expert and the conditioned PPO traffic agents (top to bottom) on the same scenario at four time steps over T=3\,\mathrm{s} (left to right). Agents are shown with their past trajectories.

Figure 13 : Qualitative comparison of traffic agent models. We visualize rollouts of IDM, PPO, SMART, expert and the conditioned PPO traffic agents (top to bottom) on the same scenario at four time steps over T=3\,\mathrm{s} (left to right). Agents are shown with their past trajectories.

Figure 3 : Planner performance across traffic agents and benchmark splits. Each radar plot shows the six core metrics, At-Fault Collision, Off-Road, Goal completion, Comfort, Center Alignment, and Lane Alignment, for a single (planner, traffic agent) combination. Rows correspond to the four planners and columns to the eight traffic agents. All axes are oriented such that the outer ring corresponds to the best observed value, with At-Fault Collision and Off-Road inverted. Each metric is independently min-max normalized across both splits and all (planner, traffic agent) cells, so axis ranges ar

Figure 6 : Score of the PDM+PPO hybrid planner as a function of top-K on Interactive1k (left) and Random1k (right). Each red marker reports the mean score (scaled by 100) over all scenarios for one traffic agent type, and the box summarizes the resulting distribution across traffic agent types at a fixed planning horizon of 10 steps (1 s). Performance improves with increasing top-K on both splits, with diminishing returns beyond K=8 . The spread across traffic conditions is markedly larger on Interactive1k than on Random1k, reflecting the heavier reactive demands of the interactive scenarios.