Force, Impedance, and Trajectory Learning for Contact Tooling and Haptic Identification

Simultaneously Encoding Movement and sEMG-Based Stiffness for Robotic Skill Learning

“…G J (x) takes into account the effect of arm geometry in the presence of external force f 0 and gravity load τ g (x). 5and (6) suggest that the Cartesian stiffness profile depends on the joint stiffness (through the muscle activities of contraction and co-contraction), the exerted external force and gravity. In our case, for simplicity f 0 and τ g (x) are dropped within the identification of the human arm parameters as suggested in [23].…”

“…Especially, for in-contact tasks where force profiles in addition to positional profiles need to be regulated [4], PbD allows relaxing the analytical burden required for the process of human-to-robot physical skills transfer [5]. One of the challenges is to enable a robot to learn human-like behaviours with flexibility and impedance adaptation [6][7][8][9]. Especially for force-dominant tasks [10], this challenge needs to be addressed urgently.…”

Learning the Correct Robot Trajectory in Real-Time from Physical Human Interactions

“…Local modifications. When the human end user applies forces and torques, the robot can respond by locally modifying its trajectory [1,9,11,15,24,27]. Akgun et al [1] enable the human to physically adjust important keyframes along the robot's trajectory so that-the next time the robot performs the task-it moves through these waypoints.…”

“…Akgun et al [1] enable the human to physically adjust important keyframes along the robot's trajectory so that-the next time the robot performs the task-it moves through these waypoints. Haddadin et al [15] and Li et al [24] develop reaction strategies to modify the robot's next few waypoints: the robot deforms its trajectory in the direction of the human's applied disturbance [9,27] or to maintain a pre-defined interaction force. For each of these approaches, the robot updates a local segment of its trajectory (e.g., a keyframe or the next few waypoints), but the robot does not learn a new trajectory from start to goal.…”

“…One common pHRI strategy is for the robot to modify its trajectory to move in the direction of the human's correction [11,15,24,27]. Here, the robot treats the human's interactions as local information-where the human expert pushes, pulls, or twists the robot toward a better state-and the robot adjusts its current waypoint (or next few waypoints) to match the human.…”