Simulated annealing and Boltzmann machines: a stochastic approach to combinatorial optimization and neural computing
Numerical Optimization of Computer Models
Numerical Optimization of Computer Models
Machine Learning
Acquiring Visual-Motor Models for Precision Manipulation with Robot Hands
ECCV '96 Proceedings of the 4th European Conference on Computer Vision-Volume II - Volume II
Global Convergence of Genetic Algorithms: A Markov Chain Analysis
PPSN I Proceedings of the 1st Workshop on Parallel Problem Solving from Nature
Dynamic Programming
Experiments on Dextrous Manipulation without Prior Object Models
Experiments on Dextrous Manipulation without Prior Object Models
Hi-index | 0.00 |
We present a method for autonomous learning of dextrous manipulationskills with multifingered robot hands. We use heuristics derived fromobservations made on human hands to reduce the degrees of freedom ofthe task and make learning tractable. Our approach consists oflearning and storing a few basic manipulation primitives for a fewprototypical objects and then using an associative memory to obtainthe required parameters for new objects and/or manipulations. Theparameter space of the robot is searched using a modified version ofthe evolution strategy, which is robust to the noise normally presentin real-world complex robotic tasks. Given the difficulty of modelingand simulating accurately the interactions of multiple fingers and anobject, and to ensure that the learned skills are applicable in thereal world, our system does not rely on simulation; all theexperimentation is performed by a physical robot, in this case the16-degree-of-freedom Utah/MIT hand. Experimental results show thataccurate dextrous manipulation skills can be learned by the robot in ashort period of time. We also show the application of the learnedprimitives to perform an assembly task and how the primitivesgeneralize to objects that are different from those used during thelearning phase.