Nikhil Sobanbabu

I'm an Masters in Electrical and Computer Engineering Student at Carnegie Mellon University. I work with Prof. Guanya Shi at the LeCAR Lab. My current work is on reducing the sim2real gap of legged robots using real2sim techniques.

I received my Bachelors in Electrical Engineering from IIT Madras during which I interned as a Robotics Intern at DiFACTO Robotics and Automation focusing on navigation and recovery of in house AMR. I also worked as a Research Intern at Dynamic Robotics and Control Laboratory under the guidance of Prof. Quan Nguyen on Model-based oracle using preview control to guide a Reinforcement Learning policy for Agile Bipedal Locomotion. Prior to this I worked with Prof. Bharath Bhikkaji on synchronised control of mutliple Crazyflie Quadrotors to track a given trajectory, and on my Bachelor Thesis on Multi-agent RL based control guided by a reach-avoid model to solve a 3D pursuit-evasion-target scenario. Apart from this I was part of Team Anveshak ,a student run Mars Rover Team of IIT Madras which participates in University Rover Challenge at Mars Research Desert Station,Utah and worked as an Embedded Systems and Control Engineer, later getting promoted as Team lead for the period July 2022-February 2023.

Email  /  CV  /  Linkedin  /  Github

Research

I am interested in Model-based Optimal Control, Reinforcement Learning and Motion planning with specific focus on legged and space robotics. I am also open to explore locomanipulation and multi-agent control.

[Webpage|Arxiv]

A Multi-modal policy guided by a Model-based Oracle, enabling agile parkour-like behaviours on a biped over a structured terrain.

[Arxiv]

Solving uni-object locomanipulation tasks using a multi-modal policy guided by a oracle designed using a hybrid automata perpespective.

Optimal Strategies found for a 1vn game by solving the HJI-PDE equation for a 1vn game with a reach-avoid objective.

[Thesis Report]

Coordinated autonomous control of multi-agent quadrotors in a regular polygon formation with varied orientations using the in-built mellinger controller.

[Report] | [Poster]

Autonomous control of Manipulator using inverse kinematics and ROS moveit.

Autonomous navigation of an Agricultural robot to identify and pick and place bellpepers inside a greenhouse.

[Video] | [Github]

Planner chooser behaviours encoded as latent modes using monte-carlo roll outs to be robust against perturbations generating emergent transitions between modes/behaviors.

[Results Video]

Simulation environment for the single attacker, singe target, multiple defender pursuit evation differntial game.

Swing-up is done using a control from an energy based Lyapunov function. After a reaching an appropriate angle, pole-placement based stabilisation kicks in.

[Report] | [Presentation video] | [Github]

Problem Statement: Optimal pick and place of multiple objects to a given destination with payload constraints for the mobile manipulator.
Solution: Dijkstra+Modified TSP for navigation. RRT* for manipulation.

[Report] | [Github]

Fun Mini-Project

Why WASD when you can play minecraft with ESP32

Miscellaneous

Flagship event of Shaastra, IITM's Technical festival for the year 2022. Participants convert a prototype to a minimum viable product.

[Link]

Template from jonbarron.