Safe Streaming Flow Planning by Aligning Sampling Dynamics with Execution Dynamics
SafeStreamingFlow enforces safety only on the next executed step, cutting the cost of prior full-trajectory safe planners.
The paper says earlier safe diffusion or flow planners repeatedly adjust intermediate states across an entire planned trajectory, which is slow and can push sampling away from execution dynamics. Its proposed planner instead streams decisions by integrating a learned state vector field with hierarchical state prediction. Safety constraints are applied through high order control barrier functions at the executed step. In navigation, racing, and locomotion benchmarks, the authors report lower planning latency and better safety while keeping competitive goal-reaching success. ArXiv · AI/CL/LG's note
The paper says earlier safe diffusion or flow planners repeatedly adjust intermediate states across an entire planned trajectory, which is slow and can push sampling away from execution dynamics. Its proposed planner instead streams decisions by integrating a learned state vector field with hierarchical state prediction. Safety constraints are applied through high order control barrier functions at the executed step. In navigation, racing, and locomotion benchmarks, the authors report lower planning latency and better safety while keeping competitive goal-reaching success. ArXiv · AI/CL/LG's note
score 4