Vision-Language-Action Model Integration

I integrated Vision-Language-Action (VLA) models with a UR5e robotic arm to enable natural language control of robotic manipulation tasks. The project involved deploying a complete perception and control pipeline on an NVIDIA Jetson Orin, implementing real-time motion execution in ROS + URScript, and achieving sub-200ms end-to-end latency from command input to motion initiation. Key challenges included debugging singularities, handling workspace edge cases, calibrating coordinate frames, and optimizing inference performance for autonomous manipulation.

Autonomous CubeSat Grasping System

At Redwire Space, I developed an autonomous robotic grasping system for CubeSat manipulation using a UR5 arm and Intel RealSense L515. Built a complete ROS-based motion planning pipeline integrating ArUco marker detection, point cloud processing, and MoveIt for collision-aware trajectory planning.

Multi-Target Space Debris Removal with APF-MPC

Developed an innovative algorithm combining Artificial Potential Fields with Model Predictive Control for autonomous multi-target space debris collection. Created a novel polygonal queue ordering system to optimize debris collection sequence, achieving 87.5% reduction in path length and 97% reduction in fuel consumption compared to baseline methods. Implemented complete trajectory optimization in MATLAB with obstacle avoidance, thruster saturation handling, and real-time adaptive control for efficient single-trip debris removal missions supporting future orbital sustainability.

Interests

  • Autonomous Robots
  • VLA Models
  • Robot Operating System
  • Computer Vision
  • Motion Planning
  • Reinforcement Learning
  • Kinematics
  • Controls
  • 3D Modeling
  • Everything else that comes with a robot 😄

About

Thanks for stopping by! I’m Manas, a robotics engineer with a Masters degree in Robotics and Autonomous Systems for Boston University. I’ve had the privilege of working on cool projects, from configuring various robotic platforms to implementing advanced path planning and motion control algorithms. My core interests include robotic perception, motion planning, and optimization techniques. Whether it’s developing reinforcement learning models for obstacle avoidance or integrating sensors for precise object manipulation, I enjoy building autonomy for robots from the ground up. When I’m not immersed in robotics, you’ll likely find me exploring national parks on a hike or stargazing in some of the world’s darkest and most serene locations.

-Manas

Want to know more? Check out my blog!

Contact

Send me a message, always excited to meet someone new 😊