AI RESEARCH

Parallel Differentiable Reachability for Learning and Planning with Certified Neural Dynamics and Controllers

arXiv CS.AI

ArXi:2605.25346v1 Announce Type: cross Neural network (NN) dynamics models and control policies achieve strong performance in robotics, but providing sound guarantees under uncertainty remains difficult, especially for closed-loop NN systems. Existing reachability tools provide formal over-approximations, yet are often non-differentiable, overly conservative, or too slow for modern learning and online planning pipelines.