Problem
Research questions and friction points this paper is trying to address.
end-to-end autonomous parking
runtime supervision
failure correction
safety shield
Timed Rule-Based Supervision of an End-to-End Autonomous Parking Policy
🤖 AI Summary
This study addresses recurrent failure modes in end-to-end autonomous parking policies within fixed scenarios, including boundary violations, braking delays, and control oscillations. We propose a rule-based parameterized safety shield incorporating temporal thresholds. Built upon the CARLA simulation platform and a Vision Transformer architecture, this method employs a manually calibrated rule engine—governed by velocity, position, and duration criteria—to intervene in and rectify control outputs. Crucially, it effectively mitigates specific failure patterns of existing policies without requiring retraining. Closed-loop evaluations demonstrate that the target parking success rate improves from 85.16% to 97.66%, with an average positional error of merely 0.21 meters, thereby significantly enhancing system safety and robustness.
📝 Abstract
We study whether a manually specified runtime supervisor can correct recurring failures of an existing end-to-end parking policy in a fixed CARLA parking lot. The vision-based Transformer architecture is inherited from Yang et al.; our contribution is a timed, rule-based Parametric Safety Shield (PSS) applied to its control outputs. The PSS uses hand-calibrated speed, position, and duration thresholds to intervene in observed failure modes, including boundary exits, delayed braking, and stalled or oscillatory control. In the reported closed-loop evaluation, 16 held-out target slots and six initial poses are each evaluated in four rounds (384 attempts per configuration). Target success increases from 327/384 (85.16%) for the retrained policy to 375/384 (97.66%) with the PSS; mean position and orientation errors among successful attempts are 0.21m and 0.33 degrees. These results show an improvement within this simulator setup. The repeated attempts share one map, vehicle, and sensor configuration, and the PSS uses simulator world coordinates; thus the results do not establish generalization to other lots or real vehicles, or a formal safety guarantee.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Parametric Safety Shield
End-to-End Autonomous Parking
Rule-Based Supervision
Vision Transformer
CARLA Simulation
🔎 Similar Papers
No similar papers found.
