Can robots find smooth paths through tight spaces?

Robot motion planning in cramped spaces with obstacles remains difficult because feasible paths are fragmented and constrained. This work uses natural functional gradients—optimization updates defined directly in function space rather than on time-discretized waypoints—to find smooth, collision-free trajectories. The method works with only black-box trajectory evaluations, meaning it doesn't need analytical gradients from contact-rich simulators. Tests on constrained manipulation tasks show smoother, more feasible paths than existing planners in narrow-clearance environments.