Kinematic constraint method improves robotic articulation evaluation
Kinematic Constraint-Referenced Evaluation for Articulated Object Manipulation
RoboticsArtificial Intelligence
Summary
Robots often try to open or move objects like doors or drawers, but usually we only say if they succeeded or failed without checking if they moved the object correctly. The authors created a method that breaks down the robot’s movement into parts that actually follow the object's joint motion and parts that waste effort by moving off the joint’s allowed path. Their method uses a model of the object and a simple controller, so it doesn’t need past examples or training. Tests on many objects showed that this method can better tell how well the robot is moving the parts relative to the object's joints, even if the exact joint angles are a bit off.
What this means in practice
- •For robotic automation engineers: Evaluate articulated object manipulation more precisely by separating correct joint motions from wasted movements without needing demonstrations.
- •For robotic software developers: Improve robotic control policies by diagnosing specific kinematic errors in manipulation of objects with joints during simulation and training.
Authors
Baptiste Bonin, Caro Strickland, Audrey Durand
Abstract
Manipulation policies are evaluated predominantly by binary task success, which cannot tell whether the commanded gesture is the one the object's kinematics admit. We introduce Constraint-Referenced Strategy (CRS), which projects the commanded end-effector twist onto the object's joint axis, separating progress along the joint from off-constraint waste. CRS requires a joint model and a joint-following scripted controller derived from it, but no recorded demonstration or learned judge. On 74 simulated articulated objects, success registers three transfer failures as lower rates but attributes none, while the two CRS channels give each a distinct signature. A door-trained behavior-cloning policy opens drawers in 59 of 110 episodes, and in these passes it commands about $40\times$ the off-axis rotation of a drawer-trained policy's passes ($1.27$ vs.\ $0.032$~rad/s). Relative to a scripted controller scored with the same axis, CRS verdicts are unchanged under joint-axis errors up to $30^\circ$; absolute thresholds tolerate about $8^\circ$.