Ganga Nair B

Hi — I’m Ganga Nair B

Robotics engineer • Controls • Reinforcement learning

I am a robotics engineer with a background in mechanical systems and a current M.Tech student at IISc, focusing on reinforcement learning and control for quadruped locomotion.

See projects Download resume

Contacts and Links

Ganga Nair B

Current work and interests

I am an M.Tech student at IISc, Bangalore. My research emphasis is on reinforcement learning–based control for quadruped locomotion, with a focus on integrating planning and learning to improve adaptability. I am currently exploring adaptive gait strategies for quadrupeds to achieve reward-driven performance. I am keen to collaborate on:

News & Highlights

Presenting at Indian Control Conference
Presenting at Indian Control Conference 2025
Presenting Poster at Humanoids 2025
Presenting Poster at Humanoids 2025
Receiving award
Receiving Kanako Miura Award

Robotics Projects

Deployment Diagram

Real-time gait adaptation using RL and MPC

Model-free RL policies for quadruped locomotion often converge to a single gait, limiting adaptability and efficiency. We propose a control framework for real-time gait adaptation using Model Predictive Path Integral (MPPI) control and a Dreamer-based module. Our approach jointly optimizes actions and continuous gait parameters to enable smooth transitions, velocity tracking, and energy-efficient locomotion. Simulation results on the Unitree Go1 show up to 40% lower energy consumption compared to fixed-gait RL policies, while maintaining accurate tracking and stable transitions across a range of target speeds. Website → | Paper →

STRIDE structured generative dynamics model

STRIDE: Physics-Guided Generative Dynamics for Robots

STRIDE is a dynamics learning framework that helps predict future states of robotic systems from current observations. Such dynamics predictions are crucial but challenging for legged robots as they operate in contact-rich environments which deal with a lot of non-conservative forces. The model combines a Lagrangian Neural Network (LNN) to capture structured rigid-body mechanics with a generative residual model based on Conditional Flow Matching (CFM) to represent stochastic interaction effects such as friction and impacts.

This structured–stochastic decomposition improves long-horizon prediction stability and contact modeling, enabling more reliable model predictive control (MPPI) for legged robots. The approach is validated on the Unitree Go1 quadruped and Unitree G1 humanoid in simulation and hardware experiments.

Paper → Website →

Safe optimal control using physics-informed learning
Physics-informed machine learning framework for safety-critical control

Physics-Informed Learning for Safe and Optimal Control

Developed a physics-informed machine learning (PIML) framework to jointly optimize safety and performance for autonomous systems. The method formulates control as a state-constrained optimal control problem and learns the solution of the Hamilton–Jacobi–Bellman (HJB) PDE using neural networks, enabling scalable control for high-dimensional nonlinear systems.

Introduced a conformal prediction–based verification scheme that quantifies performance degradation due to learning errors. The framework was validated on several robotic tasks including autonomous navigation, pursuit-evasion, and multi-agent coordination, demonstrating improved safety and lower cost compared to constrained RL and MPC baselines.

CPG-RL control framework

Quadruped Locomotion using CPG-RL - Study on Morphology-Aware Optimisation

All legged animals rely on rhythmic primitives to achieve robust and efficient locomotion across diverse terrains. Inspired by this principle, we explore a hybrid locomotion framework that combines Central Pattern Generators (CPGs) with Bayesian Learning based optimisation techniques to achieve adaptive and energy-efficient quadruped walking.

In this work, CPGs encode the underlying rhythmic structure. The parameters of the CPGs, including frequency, amplitude, and phase parameters, are optimised conditioned on the morphology using Conditional Bayesian Optimization. We us a Tree Parzen Estimator to model the performance landscape and guide the search.

The framework is being developed in collaboration with Prof. Jun Morimoto (Kyoto University) ans tested on hardware.

Model-Based Control with Online Dynamics Learning

Explored control of nonlinear systems with unknown dynamics by integrating iLQR with an online-trained neural network for dynamics estimation.

Code → |

Safe swarm navigation using control barrier functions
Safe Swarm Navigation using Control Barrier Functions

Designed a control framework for multi-quadrotor swarms using Control Barrier Functions, combined with priority-based ordering to enable collision-free traversal through constrained environments.

Comparison b/w Kimera and RTAB
Comparison b/w Kimera and RTAB

3D point cloud reconstruction using SLAM and subsequent semantic segmentation. Comparison to RTAB-Map for accuracy.

Project 3
Formation control

Leader–follower formation control for heterogeneous multi-agent systems with real-time leader tracking.

Projects in Mechanical Engineering

Dynamic vibration absorber with MR Damper

Dynamic vibration absorber with MR Damper

Designed, fabricated, and tested a dynamic vibration absorber using a Magneto-Rheological fluid based damper. Included mathematical modelling, CAD modelling and simulation, optimisation, fabrication of test setup and experimental validation.

Repo → | Report →

Suspension for Electric All-Terrain-Vehicle

Suspension for Electric All-Terrain-Vehicle

Designed, simulated and fabricated an Electric All terrain Vehicle for Electric BAJA 2025. My focus was on suspension design and optimization for rough terrain including shock absorption, stability and steering performance, without compromising on manufacturability and cost-effectiveness.

Patents & Publications

  • Ganga Nair B, Prakrut Kotecha, Shishir Kolathaya, "STRIDE: Structured Lagrangian and Stochastic Residual Dynamics via Flow Matching" Submitted to 2026 IEEE International Conference on Robotics & Automation (ICRA)Workshop on the Path Towards Generalizable Contact-Rich Robotics: Control and Representation. Paper Link →, 2026.
    Paper | Website
  • Manan Tayal, et al. "A Robust Physics-Informed Machine Learning Framework for Safe and Optimal Control of Autonomous Systems" Submitted to International Journal of Robotics Research (IJRR)., 2026.
  • Prakrut Kotecha, Ganga Nair B, Shishir Kolathaya, "STRIDE: Structured Lagrangian and Stochastic Residual Dynamics via Flow Matching" Submitted to 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)., 2026.
    Paper | Website
  • Ganga Nair B, et al. "Real-time gait adaptation for quadruped robots using RL and MPC." Accepted at the Eleventh Indian Control Conference (ICC), 2025.
    Paper | Presentation
  • Ganga Nair B, et al. "Real-time gait adaptation for quadruped robots using RL and MPC." Presented at 2025 IEEE-RAS 24th International Conference on Humanoid Robots, Seoul, Korea..
    Website
  • Ganga Nair B, et al. "Modelling and Simulation of Magneto-Rheological Fluid in a Damper Using COMSOL" Advances in Manufacturing, Automation, Design and Energy Technologies, ICoFT 2020, Lecture Notes in Mechanical Engineering, Springer, Singapore.
    Paper
  • Patent: Jagadeesha T., Ganga N.B., et al., Magneto-rheological Fluid-Based Dynamic Vibration Absorber, Indian Patent Application No. 202241036483, filed Jun 24, 2022. Status: Patent pending.

Work Experience

ExxonMobil logo

Project engineer - ExxonMobil Services and Technology Pvt Ltd

July 2022 – July 2024

  • Managing brownfield refinery projects for ExxonMobil Singapore Refinery, focused on Engineering management.
  • Recognized within the company thrice for bringing innovation into projects to optimize cost and schedule performance.
  • Spearheaded EM’s India safety program initiatives, working with multi-disciplinary teams across different regions.

Awards & Acknowledgements