Smooth, Repeatable, and Highly Dynamic In-Hand Manipulation via Decentralized Adaptable Multi-Actuator Trajectory Planning

Abstract

Dexterous in-hand manipulation is becoming increasingly important as robotic systems evolve toward agile, general-purpose automation. This paper presents a decentralized, grid-based trajectory planning approach for in-hand manipulation that coordinates the eleven degrees of freedom of a fully pneumatically actuated anthropomorphic robotic hand. The planner uses uniform time discretization and continuously differentiable second-order point-to-point trajectories in position and velocity, which allows intuitive manual tuning of coordinated multi-actuator motions. Experimental validation on two in-hand manipulation tasks demonstrates smooth, highly dynamic, and repeatable execution of complex ball rotations, despite the lack of sensing for the soft finger actuators and object pose estimation. The results are achieved by combining feedforward control of the soft finger actuators with feedback control of the rigid palm actuators considering friction compensation. The proposed trajectory planning approach is generalizable and transferable to robots with parallel kinematics and partially observed states.