A majority of microrobots are constructed using compliant materials that are difficult to model analytically, limiting the utility of traditional model-based controllers . We propose anovel framework residual model learning (RML) that leverages approximate modelsto substantially reduce the sample complexity associated with learning anaccurate robot model . We show that using RML, we can learn a model of theHarvard Ambulatory MicroRobot (HAMR) using just 12 seconds of passivelycollected interaction data . The learned model is accurate enough to beleveraged as “proxy-simulator” for learning walking and turning behaviors using model-free reinforcement learning algorithms .

Author(s) : Joshua Gruenstein, Tao Chen, Neel Doshi, Pulkit Agrawal

Links : PDF - Abstract

Code :
Coursera

Keywords : model - learning - microrobot - rml - residual -

Leave a Reply

Your email address will not be published. Required fields are marked *