Planning Motions with Intention Presented by Yan Ke

Planning Motions with Intention Presented by: Yan Ke NUS CS 5247

Problem Specification p Task: Generate motions for human or robot arms to complete manipulation tasks. p Goal: Find a collision-free path in configuration space. p Tool: Inverse kinematics algorithm. p Usage: Computer animation. NUS CS 5247

NUS CS 5247

Difficulties p Motion with Intention: Human and robot arms are moving with the intention of completing some task. p Restriction: Neurophysiology. p Grasping and Regrasping: Posture; multiple arms. p PSPACE-hard NUS CS 5247

Section 1 Manipulation Planning Overview NUS CS 5247

Inputs p Geometry of the arms p Movable object p Obstacles together with their locations p Initial and goal configuration NUS CS 5247

The Stable Space and Grasp Space p Stable space: The set of all configurations where the movable object M is statically stable. p Grasp Space: Arms grasping M and moving it stably. p Grasp Space Stable Space the Configuration Space NUS CS 5247 Free Space of

Transit Paths and Transfer Paths p Transit Paths: Arms motions that do not move M p Transfer Paths: Arms motions that move M NUS CS 5247

Planning Result NUS CS 5247

Section 2 Generating Transfer and Transit Paths NUS CS 5247

Overview p The entire manipulation planning can be accomplished by following: ① Generate a series of subtasks to achieve the goal configuration. ② Plan a transit or transfer path for each subtasks. p Assumption: Transit tasks can be completed by transit paths; transfer tasks can be completed by transfer paths. NUS CS 5247

Generating Transfer Tasks p Grasp set: All various possible grasps for a certain M. p Grasp assignment: A pair associates with an element in grasp set and an identity of the grasping arm(s). p We first generate the path for M moving alone. p Secondly, we attach each configuration of M with a list of grasp assignment. NUS CS 5247

Generating Transfer Tasks p The attached list of grasp assignments are obtained by pruning out those no longer possible in the new configuration from the previous configuration. p If somehow we found the list of grasp assignment is empty, then a regrasping is necessary here. p We solve this problem by resetting the list, find all possible of grasp again, and associate them with arm(s). NUS CS 5247

Assumptions p An arm can attain a grasp with a finite set of different postures. p All arms not involved in the task is placed elsewhere without blocking the motions of working arms. p If M requires two arms to move, any one of them alone, can hold M stably to allow the other one to move in a transit path. NUS CS 5247

Result p A motion planning path for M, each configuration is attached with a list of grasp assignment. p The path is partitioned into several subpaths by regrasping. p Each subpath is a transfer task. p It does not guarantee to find the best path. NUS CS 5247

Generating Transit Paths p Transit paths are the paths moving the arms. p Connect the initial configuration to the first grasp assignment of the first transfer task. p Connect grasp assignments between different transfer tasks. p Connect the last grasp assignment to the goal configuartion. NUS CS 5247

Section 3 Human-Arm Kinematics NUS CS 5247

Neurophysiology p Goal: Determine the arm posture for a human arm given the position and orientation of its hand. p Two experimental result: ① Arm and wrist posture are for the most part independent of each other. ② Arm posture for pointing is mainly determined by an ST model. p ST model: Can determine shoulder and elbow joint angles given the position of hand. NUS CS 5247

Arm Posture p Ø NUS CS 5247 What do we have? R, ψ, X What do we want? θ, β, α, η

Inverse Kinematics Algorithm NUS CS 5247

Illegal Posture Adjustment NUS CS 5247 p Claim: εis the only one to violate its limits. p Solution: Decrease Φ. p Result: wrist position unchanged when Φ decrease.

Experimental Result NUS CS 5247

Experimental Result p Working environment: C and UNIX. p Time used: three and a half minutes. p Identify the transfer tasks: one and a half minutes. p Different grasp assignments in total: 2600. NUS CS 5247

Conclusion p A novel approach to solve the multi-arm manipulation planning problem. p Computation time is unbounded. If no path exist, the algorithm may run forever. p Aim to create a task-level animation package for human motions. NUS CS 5247