Self-Collision Detection and Prevention for Humanoid Robots

James Kuffner

Abstract

This is joint work with Satoshi Kagami, Koichi Nishiwaki, Masayuki Inaba, and Hirochika Inoue. (Tokyo Univ.)
In order for humanoid robots to become practical, they must be able to operate safely and reliably. "Self-collisions" occur when one or more of the links of a robot collide. Self-collisions can result in damage to the robot itself, or through a loss of balance or control, cause human injury or damage to its surrounding environment. Thus, detecting and avoiding self-collisions is fundamental to the development of humanoid systems which can be safely operated in human environments.
In this talk, I will present some ongoing work on developing efficient geometric approaches to detecting link interference suitable for complex articulated robots such as humanoids. Software that I developed for the University of Tokyo "H7" humanoid utilizes feature-based minimum distance determination methods for convex polyhedra in order to conservatively guarantee that a given trajectory is free of self-collision. Threshold values can be set on the allowable minimum distance between links in order to provide a safety margin that accounts for errors in modeling and control. Full body trajectories can be checked in advance for potentially self-colliding postures prior to being executed on the robot.
In the current implementation, the minimum distances between all possible relevant body link pairs for a 30 DOF humanoid (435 pairs) can be calculated in approximately 2.5 milliseconds on average on a typical PC. Although we have focused on detecting self-collisions for biped humanoids, the technique can generally be applied to any robot with articulated appendages (arms or legs). It is also applicable to detecting inter-robot collisions for multiple manipulators which share a common workspace (e.g. crowded factory workcells).
I will be showing several video clips of both simulations and experiments using actual humanoid robot hardware. Finally, I will briefly discuss some of the exciting prospects for humanoid robotics as companies, such as Honda and Sony, have announced some new humanoid robot prototypes.

Presentation Slides

Pdf (1.5Mb)

 

Speaker Appointments

For appointments, please contact James Kuffner (kuffner@cs.cmu.edu).