Regrasp Planning: Reorienting an Object via Pick-and-Place
Interactive regrasp-planning simulator: a robot gripper with a limited wrist-rotation range must reorient a cube to a target face-up pose. Watch it search a stable-pose graph with breadth-first search and execute a multi-step pick / rotate / place sequence when a single grasp can't reach the target directly.
Dexterous manipulation is not always about fingers rolling an object in the palm — often the wrist simply cannot turn far enough in one motion to reach a target orientation, and the arm has to set the object down and pick it back up from a different angle instead. This simulator models that as a graph-search problem: six stable "face-up" poses of a cube become nodes, and a direct grasp-and-rotate edge connects two poses only when the required wrist rotation is within the gripper's limit. Pick a target face and the arm runs a real breadth-first search over that graph to find the shortest regrasp sequence, then physically executes it step by step — approach, grasp, lift, rotate, lower, release — while live readouts track the plan length, how many graph nodes the search touched, and the cumulative wrist rotation the arm actually performed. Tighten the wrist-rotation-limit slider below 180° and watch opposite-face reorientations, which used to take one grasp, split into a two-hop regrasp through an adjacent pose.
A robot gripper with a limited wrist-rotation range must reorient a cube to a target face-up pose, running breadth-first search over a stable-pose graph and executing a multi-step pick / rotate / place sequence whenever a single grasp can't reach the target directly.
3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install