TrajEdit is a hierarchical trajectory planner designed for pursuit-evasion scenarios, where a dynamic obstacle actively chases the ego agent. The system follows the current plan under nominal conditions and adapts only the remaining portion of that plan as the pursuer approaches. This avoids the computational cost of regenerating trajectories from scratch at every step.
The core editing step relies on FlowEdit, a flow-matching method for trajectory adaptation. When the adversarial agent comes within a safety threshold, TrajEdit falls back to a control barrier function safety filter. According to the paper, this approach reduces computation time while maintaining a success rate comparable to full replanning baselines.
The authors evaluate TrajEdit in 2D pursuit-evasion environments with static obstacles and an adversarial pursuer. They report high success rates with low capture and collision rates, along with computational time savings relative to full replanning. The paper is available on arXiv under identifier 2610.24256.