Shivam Sood

Accepted at IROS 2024 [paper][video]

Optimal Control for legged robots has gone through a paradigm shift from position-based to torque-based control, owing to the latter’s compliant and robust nature. In parallel to this shift, the community has also turned to Deep Reinforcement Learning (DRL) as a promising approach to directly learn locomotion policies for complex real-life tasks. However, most end-to-end DRL approaches still operate in position space, mainly because learning in torque space is often sample-inefficient and does not consistently converge to natural gaits. To address these challenges, we introduce Decaying Action Priors (DecAP), a novel three-stage framework to learn and deploy torque policies for legged locomotion.

Accepted at ICRA 2023 [paper][website]

This work presents a simple linear policy for direct force control for quadrupedal robot locomotion. The motivation is that force control is essential for highly dynamic and agile motions. Unlike the majority of the existing works that use complex nonlinear function approximators to represent the RL policy or model predictive control (MPC) methods with many optimization variables in the order of hundred, our controller uses a simple linear function approximator to represent policy.

Accepted at ICCR 2022 [paper][code]

We present a multiple waypoint path planner and controller stack for navigation in unknown indoor environments where waypoints include the goal along with the intermediary points that the robot must traverse before reaching the goal. Our approach makes use of a global planner (to find the next best waypoint at any instant), a local planner (to plan the path to a specific waypoint) and an adaptive Model Predictive Control strategy (for robust system control and faster maneuvers). We evaluate our algorithm on a set of randomly generated obstacle maps, intermediate waypoints and start-goal pairs, with results indicating significant reduction in computational costs, with high accuracies and robust control.

MARMot Lab, NUS [website]

This project is concerned with training a Proximal Policy Optimization based parallel DRL control policy for control of a hexapod robot lovingly named Yuna (based on a game character with different eye colors since that’s what Yuna’s LEDs look like :p)

Stoch Lab, IISc Bangalore [website]

The gaol of the project was to develop a Representation Free Model Predictive Control for a in-house quadruped robot named Stoch3. This included solving for optimal Ground Reaction Forces based on a Single Rigid Body dynamics model while following gaits programmed using a finite state machine. The model used rotation matrices directly to get rid of the issues with euler angles (gimbal lock) and with quaternions(unwinding).

AGV, IIT Kharagpur [LQR Code] [Geometric Controls]

This project was aimed at benchmarking various control algorithms for autononomous ground vehciles. This included algorithms like Model-Predictive-Control, Linear Quadratic Regulator, Pure Pursuit and Stanley.

IIT Kharagpur and DRDO [Presentation]

This competetion was targeted at co-ordinating a UAV and an Unmanned Snow clearing vehicle for navigation in a hilly snow covered terrain and the UGV's autonomous traversal. We developed a Non-Linear Model Predictive Control based controller for the UGV and a path planning algorithm for the UAV. We won the gold medal in the competetion.

IIT Kharagpur and Drona Aviation

This goal of this challenge was to develop a vision based state feedback control for an indoor multi-drone system handling socket communication with the flight controller without the use of ROS. I led the team for this competetion and we won a gold.

MARMot Lab [code]

This code includes various implementation of a model predictive control including multiple and single shooting methods. It includes both implementations for point tracking as well as trajectory tracking for mobile robots.