A feasibility-driven approach to control-limited DDP

Abstract Differential dynamic programming (DDP) is a direct single shooting method for trajectory optimization. Its efficiency derives from the exploitation of temporal structure (inherent to optimal control problems) and explicit roll-out/integration system dynamics. However, it suffers numerical instability and, when compared multiple methods, has limited initialization options (allows controls, but not states) lacks proper handling constraints. In this work, we tackle these issues with feasibility-driven approach that regulates feasibility during optimization ensures limits. Our search emulates resolution problem only dynamics We show our (named Box-FDDP ) better convergence than Box-DDP $$^+$$ + (a method), its rate runtime performance are competitive state-of-the-art transcription formulations solved using interior point active set algorithms available in Knitro . further decreases error monotonically—as nonlinear algorithms. demonstrate benefits by generating complex athletic motions quadruped humanoid robots. Finally, highlight suitable model predictive legged