Task-specific acceleration-domain trajectory optimization for smooth and energy-efficient industrial manipulator motion

Serial manipulators performing partially constrained tasks possess trajectory-level redundancy that can be exploited to reduce jerk and energy use. We propose a task-specific trajectory optimization framework using a latent acceleration panel representation. By parameterizing joint acceleration in a Legendre basis, velocity and position are recovered analytically through anti-derivatives without finite-difference approximation. The framework is evaluated on a UR5e manipulator across four tasks covering single-target reaching, orientation-constrained reaching, multi-waypoint inspection, and payload-aware energy minimization. Against conventional baselines, it reduces integrated squared jerk by 48%–75% and shortens trajectories while maintaining sub-millimeter accuracy. With a 3.0 kg centered payload, joint mechanical work decreases by 29% and gross electric energy by 42% on hardware. The framework thus enables smoother, more energy-efficient manipulator motion.