Current Projects

Soft-walk: Silent Legged Locomotion

Jan 2023 - Present

Given the ability to train and deploy a reinforcement learned (RL) controller on the UniTree Go1 robot, in a simulation to real process, can we shape the reward such that less sound is emitted when walking/trotting/running? Legged-robots have been shown to be extremely useful in situation where traversing harsh terrain is required. The kind of terrain, where the robot might be expected to jump or climb over/about obstacles. However, these kinds of robots do not tend to traverse around in a way that seems natural to spectators, particularly when comparing them to what is seen in nature. Even when walking around on flat ground, legged-robots tend to be quite loud, pounding into the ground. This is not only audibly obtrusive to onlookers, but also wasteful of energy. We are seeking to reduce this effect seen in legged robots through the use of a RL learned control strategy.  (paper in progress)

Multi-task Morphology Optimization

July 2022 - Present

Expanding on the work I completed in my graduate studies in addition to other work completed in this general field, we seek to discover a method for quantifying a morphologies effectiveness on unforeseen tasks. To do this, we utilize competence-based unsupervised reinforcement learning (URL) methods to assign fitness values within a morphology optimization loop. Initial findings show, that compared to alternative techniques, our method tends to rank morphologies in a way where the fitness assigned better correlates to final mulit-task performance reward. (paper in progress)

Trans-modal Reinforcement Learning

Oct 2022 - Present

Info coming.