DRAFTO: Decoupled Reduced-space and Adaptive Feasibility-repair Trajectory Optimization for Robotic Manipulators

A team of researchers led by Yichang Feng, Xiao Liang, and Minghui Zheng has published a new paper on arXiv introducing DRAFTO (Decoupled Reduced-space and Adaptive Feasibility-repair Trajectory Optimization). This novel algorithm is designed specifically for planning smooth, safe, and feasible paths for robotic arms (manipulators). Its core innovation is a decoupled two-phase approach: it uses a reduced-space Gauss-Newton descent for the primary optimization iterations, which is computationally efficient, and employs constrained quadratic programming specifically to handle initialization and repair terminal joint-limit feasibility. This separation aims to reduce the computational burden of repeated full constrained optimizations.

The method constructs an objective function that accounts for trajectory smoothness, safety, joint limits, and task requirements. It uses a hinge-squared penalty for inequality constraints and a two-phase acceptance rule with a non-monotone policy to ensure the optimization can converge from poor initial guesses (globalizability). In comprehensive benchmark tests, DRAFTO was validated against a suite of established optimization-based planners (CHOMP, TrajOpt, GPMP2, FACTO) and sampling-based planners (RRT-Connect, RRT*, PRM). The results demonstrated its high efficiency and reliability across diverse scenarios. A key experiment showing a robot successfully grabbing an object from a drawer highlights its potential for implementation in complex, real-world manipulation tasks where precise, collision-free motion is critical.