Skip to content
View Shakthi-Bala's full-sized avatar
🎯
Focusing
🎯
Focusing

Organizations

@pearwpi

Block or report Shakthi-Bala

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Shakthi-Bala/README.md

Shakthi Bala

MS Robotics Engineering · WPI · GPA 4.0 Perception · Manipulation · Aerial Robotics · Massachusetts, US


About

I build end-to-end robotic systems — from sensor fusion and learned perception to motion planning, control, and deployment on constrained hardware.

Directed researcher at PeAR Lab (WPI) under Prof. Nitin Sanket, developing lightweight multimodal perception for flapping-wing MAVs that works in snow, smoke, and darkness. Previously at MER Lab (WPI) and DManD Lab (SUTD).


Research

ThermalUSFusionNet — PeAR Lab, WPI · Jan 2026 – Present · Advisor: Prof. Nitin Sanket

  • Designed a 24 g, 1.2 mW ultrasound + thermal stack for a 148 g flapping-wing MAV — replacing LiDAR/RGB-D in degraded conditions
  • Built ThermalUSFusionNet: FiLM-conditioned encoder–decoder for per-pixel metric depth + uncertainty up to 7 m
  • Deployed full perception + RL navigation stack on Raspberry Pi Zero 2W at 18 Hz across 10+ indoor/outdoor sites
  • Robust on materials that defeat RGB-D: transparent acrylic, aluminum foil, matte black surfaces

Ensemble Antipodal Grasping — MER Lab, WPI · Aug – Dec 2025 · Advisor: Prof. Berk Calli

  • Modular benchmarking framework across 1000+ grasp attempts under noisy perception/execution
  • Ensemble grasp selection combining geometry planners + learned quality networks → +7% grasp success
  • Built failure analysis tools separating perception vs. execution failures

Projects

Project Stack Result
DeepVIO — FiLM Visual-Inertial Odometry PyTorch · ResNet-18 · BiLSTM 92% test loss reduction · 1.5 m ATE
Hybrid IK–RL Controller — UR16e MoveIt2 · PPO · Isaac Sim +15% success in cluttered scenes
CBF Quadrotor Control — Full dynamics C3BF · Geometric Control · Isaac Sim Real-time collision avoidance on min-snap trajectories
RRT Aerial Planner* ROS2 · PID · Cubic Splines Full 3D planning + cascaded PID tracking
NeRF from Scratch PyTorch · Volume Rendering Lego/Ship benchmarks + real drone scenes
Structure from Motion NumPy · SIFT · RANSAC · Bundle Adj. Reconstructed WPI Unity Hall from 5 photos
Stereo MSCKF VIO Python · EuRoC Validated vs mm-accurate Vicon ground truth
ICRA SimSort MATLAB · YOLOv4 · IK 🥉 2nd Runner-Up, ICRA 2025

Stack

Languages · Python · C++17/20 · MATLAB Robotics · ROS2 · MoveIt2 · Nav2 · PX4 · MAVSDK Simulation · Isaac Sim · Gazebo · PyBullet Learning · PyTorch · PPO · FiLM · NeRF · 3DGS Perception · OpenCV · Open3D · PCL · ZED SDK Hardware · UR16e · Crazyflie · RealSense · ZED2i · RPi Zero 2W


Highlights

  • 4.0 GPA · MS Robotics Engineering, WPI
  • 🥉 2nd Runner-Up · ICRA 2025 Simulation Sorting
  • 18 Hz perception deployed on 148 g MAV · Raspberry Pi Zero 2W
  • 1000+ grasp attempts benchmarked at MER Lab

Links

Popular repositories Loading

  1. Optimization-techniques-on-robotino3 Optimization-techniques-on-robotino3 Public

    C++ 1

  2. Coffee-serving-robot Coffee-serving-robot Public

    C++ 1

  3. RGMC_Simsort_2025 RGMC_Simsort_2025 Public

    1

  4. blender_dataset_generator blender_dataset_generator Public

    The script generates a targeted prim with multiple camera angles, multiple backgrounds, multiple orientation and occulsions

    Python 1

  5. Navigating_through_unknown-Vizflyt Navigating_through_unknown-Vizflyt Public

    This is perception + control stack with which we could navigate autonomously through gaps that are in unknown shape

    Python 1

  6. CBF_for_full_drone_dynamics CBF_for_full_drone_dynamics Public

    This repo implements a CBF safety filter on top of a Non linear Controller provided a very aggressive trajectory

    Python 1