BLAST: Biconvex Limit-Aware System Traversal

Abstract

Planning fast and dynamically feasible motions within prescribed safe regions in $C$-space is a critical requirement for many dynamical systems, including manipulators and unmanned aerial vehicles, especially when operating near their performance limits. Existing planning pipelines often compute a collision-free geometric path and subsequently time-parameterize it under kinematic limits, while neglecting the effects of the full system dynamics. As a result, the generated motions may require excessive actuator effort, increase mechanical and thermal loading during execution, and even lead to torque-limit violations. In this work, we propose BLAST, a biconvex, limit-aware path-and-timing optimization framework that couples B-spline collision-free path optimization with velocity and acceleration profiling under explicit dynamical and operational constraints. By exploiting the resulting problem structure, BLAST retains computational tractability while incorporating actuator limits and system dynamics directly into the optimization. We evaluate the proposed approach against recent state-of-the-art methods across point-mass, AAV, and manipulators, including a full 7-DoF setup. Our comprehensive results demonstrate that our anytime framework explicitly accounts for system constraints and reduces actuation energy, uses fewer control points, and substantially decreases torque-limit violations during execution.