Controlled Chaos: Stress-Testing Planners via Adversarial Perturbation Frameworks

Ransiek et al. (2023) generated adversarial trajectories via RL, but their methods lack theoretical guarantees on perturbation coverage. This research proposes a chaos-based stress-testing framework that uses Lyapunov exponents and bifurcation theory to systematically inject perturbations (e.g., sensor noise, dynamic obstacles) into planners like CB-MPC or GVI-MP (Yu & Chen, 2023). By quantifying planner resilience through perturbation phase diagrams, we could identify critical thresholds where planners fail unexpectedly—addressing gaps in robustness analysis highlighted by Zhang et al. (2022). Unlike traditional testing, this method ensures comprehensive coverage of edge cases, potentially revealing why planners like ST-RRT* succeed in some edge cases but fail in others.

References:

1. Task and Motion Planning Methods: Applications and Limitations. Kai Zhang, E. Lucet, Julien Alexandre Dit Sandretto, Selma Kchir, David Filliat (2022). International Conference on Informatics in Control, Automation and Robotics.
2. Generation of Adversarial Trajectories using Reinforcement Learning to Test Motion Planning Algorithms. Joshua Ransiek, Barbara Schütt, Adrian Hof, Eric Sax (2023). 2023 IEEE 26th International Conference on Intelligent Transportation Systems (ITSC).
3. Stochastic Motion Planning as Gaussian Variational Inference: Theory and Algorithms. Hongzhe Yu, Yongxin Chen (2023). arXiv.org.