Hi, I'm Harsh Mahesheka

Skild AI

Member of Technical Staff

MyoLab, Carnegie Mellon University

Research Intern under Dr. Vikas Kumar
Worked in geenrating human motion embodiment data from one shot in wild motions. Designed and applied biomechanically constrained optimization pipelines to refine foot trajectories and correct motion irregularities, generating improvements of up to 40% on various motion artifacts

Intelligent Robotics and Interactive Systems Lab, Arizona State University

Research Internship Under Dr. Wanxin Jin and Dr. Zhaoran Wang
Developed a hierarchical VLM-LLM architecture for inverse reinforcement learning from videos by generating reward functions. The framework uses a high-level VLM (Gemini 1.5) to analyze expert and learner videos, providing feedback that a low-level LLM (GPT-4) uses to update the reward function.

Robot Learning Lab, University of Freiburg

DAAD-WISE Research Internship Under Dr. Abhinav Valada
Developed an RL-based planner to generate base velocities for mobile manipulators, enabling them to achieve end-effector goal poses provided by the teleoperator. Utilized the planner to streamline the collection of demonstrations for kinesthetic learning.

Google Summer of Code - Project Mesa

Student Developer at Project Mesa
Enhanced the Mesa agent-based modeling framework for Multi-Agent Reinforcement Learning by creating wrappers and functions for seamless integration with reinforcement learning libraries like RLlib and Stable Baselines. Created tutorials and documentation for getting started.

Google Summer of Code - Open Robotics

Student Developer at Open Robotics
Created a tool for generating template packages for the Gazebo simulator and macros for installing and sourcing worlds, models, plugins, etc., in the Gazebo robotics simulator. These tools were designed to streamline the process of creating and deploying new packages.

Hindustan Unilever

Tech Intern at Mumbai HO
Energy Optimizer: Used machine learning and data analytics to develop a tool to serve as an end-to-end solution for optimizing parameters to enhance factory energy efficiency, generating energy insights worth 3.5 crores.

Robotics Research Center, IIIT Hyderabad

Research Internship Under Dr. K. Madhava Krishna
Worked on an RL-based Global Planner trained to reduce localization drift by avoiding symmetric or featureless areas. Also worked on grounding waypoints on annotated maps by analyzing complex language commands in indoor environments.