Siddarth dayasagar

Robotics Intern @ Spacedata Inc

About

I build humanoid locomotion and manipulation systems, training reinforcement learning policies, designing model-predictive and whole-body controllers, and deploying them from simulation onto physical hardware. My work sits at the intersection of classical control and learning-based control. On the controls side, I've built centroidal MPC running at 200 Hz with whole-body control (Pinocchio, OSQP, CasADi) for a bipedal humanoid (BHEEMA), and a three-layer autonomy stack with nonlinear MPC and an MPCC racing controller on physical F1TENTH hardware. On the learning side, I've trained RL locomotion policies in MuJoCo and Isaac Lab, and deployed them on ROS2, most recently for HANUMAN, a locomotion-and-navigation stack for the Unitree G1 humanoid built to walk on unstructured Mars-like terrain. I've also built PRANA, a flow-matching vision-action policy with a DINOv2 backbone deployed on a physical 7-DOF arm for autonomous grasp-and-place at 50 Hz. At SpaceData Inc. in Tokyo, I trained PPO-based quadruped locomotion policies in Isaac Lab and set up indoor/outdoor autonomous navigation for disaster environments. I'm an active contributor to Space Station OS, an open-source ROS2 flight-software project, where I led CI/CD restructuring and ECLSS fault-scenario work. I care most about sim-to-real: the gap between a policy that works in simulation and one that survives contact, disturbance, and uncertainty on real hardware. For me, autonomy isn't just motion, it's resilience. My goal is to build machines that think, adapt, and endure in the environments we send them into next. Currently completing an MS in Robotics at Northeastern University (Dec 2026). Open to roles in humanoid locomotion, whole-body control, and RL for robotics. Tools: ROS2 · MuJoCo · Isaac Lab/Sim · Python · C++ · PyTorch · Pinocchio · OSQP · CasADi ·

Country

United States

City

Greater Boston

Industry

Information Technology & Services

Skill

Robot Control, State Estimation, MPPI, GTSAM, PyTorch, Mujoco, Robot Manipulation, VLA model, Trajectory Optimization, Isaac sim, Modeling and Simulation, Reinforcement Learning, international space station, Gazebo, MPC, Artificial Intelligence (AI), Formal Methods, Space Systems, Optimization, Model Predictive Control

Experience

Robotics Intern

Spacedata Inc

2025-8 - 2025-12 · 5 mos

Tokyo, Japan

Engineered a PPO-based quadruped locomotion policy in Isaac Sim, improving performance by approximately 70% success rate over the company’s pretrained baseline by refining rewards, optimising hyperparameters, and implementing domain randomisation. Integrated ROS 2 Nav2 with an MPPI local planner for indoor autonomous navigation, combining global path planning with multi-sensor fusion for GPS-denied disaster-response scenarios. Built an outdoor navigation pipeline using elevation-based grid mapping and an MPC local controller, extending quadruped deployment from controlled indoor environments to unstructured multi-terrain settings. Architected Space Station OS, a modular ROS 2 autonomy framework that combined subsystems into a unified control architecture, using behaviour trees, system diagnostics, and multi-node coordination to create a scalable, fault-tolerant space station simulation platform.

Perception intern

Strider Robotics

2023-12 - 2024-7 · 8 mos

Bengaluru, Karnataka, India

Throughout my role as a perception intern, I had the opportunity to delve into various aspects of perception technology. Some of the key experiences and tasks I undertook include: - Delving into the intricacies of PCL and its utilization for processing and working with point cloud data. - Gaining practical exposure and hands-on experience with Ouster 3D lidars and the RealSense depth camera, allowing me to understand their practical applications and workings. - Engaging in the collection of diverse off-road terrain datasets, aimed at comprehending the complexities of perception challenges for legged robots in variable environments. - Taking part in the pivotal task of porting Elevation mapping into ROS2 Humble, which serves as the pivotal foundation for the robot's perception capabilities.

Robotics intern

FloMobility

2022-9 - 2023-4 · 8 mos

Bangalore Urban

Integrated a full-stack vision-based localisation pipeline using AprilTag boards, merging real-time pose estimation with the robot’s navigation stack to eliminate localisation drift. Implemented the design and deployment of Bang-Bang and PID control for physical actuators, stabilising hardware response and maximising system operational efficiency. Diagnosed and resolved sensor-to-actuator latency bottlenecks across the autonomy stack, improving pose estimation accuracy for commercial deployment.