This paper presents a data-driven control design framework to achieve robust tracking control without exploiting mathematical model of nonlinear underactuated mechanical systems (UMS). The method leverages the differential flatness property of linearized systems and online estimation and compensation of disturbances by active disturbance rejecti…
Underactuated double-pendulum (acrobot) on a cart to play with direct optimal control methods
Benchmarking SAC, TD3, and DDPG on multi-skill bipedal locomotion using progressive waypoint-based reward shaping in PyBullet. Published at IEEE ICC 2025.
This project presents a robust and energy-efficient obstacle avoidance framework for an 8-DOF bipedal robot using Deep Reinforcement Learning (Soft Actor-Critic). By tightly integrating an A* planner with a responsive local control policy, the framework successfully navigates through densely cluttered environments while maintaining stability, minim
Assignment Codes for the MIT OCW Underactuated Robotics and the Robotics Course at IITM
Underactuated N-link pendulum control from scratch: single-file PPO, MuJoCo/MJX GPU port, trajectory optimisation and MPPI. Double pendulum swing-up solved 100%; triple balance 98.7%.
Reachability guarantees for energy-based swing-up and LQR stabilization of a cart pole system
Zero-dependency TypeScript control toolkit for a double inverted pendulum (iLQR, TVLQR, NMPC, CARE/DARE Riccati solvers) capable of running at 200+ Hz.
Study notes, worked notebooks, and supplementary code for MIT 6.832
To associate your repository with the underactuated-robotics topic, visit your repo's landing page and select "manage topics."