About Me

Hello! I am Devadas Vijayan Sheela, a Master's student in Electromobility (AI & Autonomous Systems) at Friedrich-Alexander-Universität Erlangen–Nürnberg (FAU), Germany. I have over two years of research-driven experience spanning robotic manipulation, autonomous systems, and simulation environments. I currently work as a Robotics Research Assistant at the University of Technology Nuremberg (UTN), contributing to robotics and AI research using the Franka Research 3 platform.

My interests lie in AI-driven robotics, robot learning, autonomous navigation, and robotic manipulation. I am a co-author of iPack: Intuitive Bin Packing with Large Language Models and an acknowledged contributor to the real-world evaluation of the DuoBench: A Reproducible Benchmark for Bimanual Manipulation in Simulation and the Real World bimanual manipulation benchmark. My master's thesis focuses on reinforcement learning-based navigation and manipulation for a custom assistive mobile manipulator in NVIDIA Isaac Lab. Alongside my research, I maintain a personal robotics lab at home and co-run LeRobot Robot Learning Lab , an open research initiative focused on studying and reproducing state-of-the-art robot learning, imitation learning, and Vision–Language–Action (VLA) systems based on the Hugging Face LeRobot ecosystem.

Experience

Projects

Robot Learning for Autonomous Manipulation using Imitation Learning and ACT

🔗 View Project on GitHub

TOOLS & PLATFORMS: LeRobot · SO-101 · ACT · PyTorch · Hugging Face · Computer Vision · Imitation Learning

• Developed an end-to-end robot learning pipeline for autonomous pick-and-place manipulation using the LeRobot SO-101 platform.
• Trained an Action Chunking Transformer (ACT) policy with visual observations and robot joint states to learn manipulation behavior from human demonstrations.
• Validated the complete robot learning workflow including teleoperation, dataset creation, policy training, rollout execution, checkpoint evaluation, and hardware deployment.
  🔗 LeRobot Robot Learning Lab

Assistive Robot Simulation & Navigation

TOOLS & PLATFORMS: ROS 2 · ZED 2i · Isaac Sim · Nav2 Stack

• Developed URDF robot models from CAD assemblies and validated multibody dynamics in NVIDIA Isaac Sim.
• Integrated a ZED 2i depth camera for perception and depth-based navigation using ROS 2 Nav2.
• Implemented autonomous indoor navigation with real-time obstacle avoidance.

Franka Emika Manipulator & Teleoperation for VLA

TOOLS & PLATFORMS: MuJoCo · Franka Emika · Optimization · Teleoperation · VLA

• Benchmarked, evaluated, and optimized 3D bin-picking algorithms for manipulators.
• Conducted structured evaluation and teleoperation experiments with the Franka Emika robot.
• Analyzed grasp success rate, robustness, and performance metrics for 3D bin packing.

Assistive Robot Object Picking — Master's Thesis (Ongoing)

TOOLS & PLATFORMS: Isaac Lab · ROS 2 · VLA · RL

• Developing an assistive mobile manipulator in NVIDIA Isaac Sim and Isaac Lab for RL-based training.
• Performing teleoperation experiments to validate manipulation pipelines.
• Integrating Vision–Language–Action (VLA) models with autonomous navigation.
• Studying sim-to-real transfer using real hardware experiments.

Autonomous Drone Flight & Real-time Object Detection

TOOLS & PLATFORMS: YOLOv3 · Intel RealSense · OpenCV

• Implemented a YOLOv3-based object detection pipeline for autonomous drones.
• Used depth data from Intel RealSense for spatial validation.
• Validated performance using ROS bag testing and flight experiments.

Interactive Image Segmentation (SAM-2)

TOOLS & PLATFORMS: SAM-2 · Hugging Face · Human-in-the-Loop

• Implemented interactive image segmentation using SAM-2 for Franka.
• Enabled human-in-the-loop object selection for robotic grasping.
• Demonstrated segmentation on multiple grocery items.

Home Robotics Lab & Prototyping

At home, I maintain a compact yet highly functional robotics lab where I design, build, and validate robotic systems from the ground up. The lab is equipped with an Anycubic 3D printer, embedded electronics and soldering tools, and NVIDIA Jetson developer kits, enabling rapid prototyping that bridges mechanical design, embedded control, and AI-driven perception. I also train imitation learning policies (Behaviour Cloning, ACT, and diffusion-based methods) on a LeRobot SO-101 leader-follower platform and co-run Open Val Research — an open initiative for VLA model experiments.

Personal Robotics Prototyping Lab

TOOLS & ACTIVITIES: 3D Printer · Embedded Systems · Soldering · NVIDIA Jetson Dev Kit

• Prototyped custom robotic chassis through iterative 3D design and printing.
• Configured NVIDIA Jetson platforms for embedded AI and robotics workloads.
• Developed and integrated microcontroller-based subsystems through hardware programming and soldering.

Embedded Electronics & Soldering Experiments

• Designed and assembled embedded circuits using ESP32, TB6612 motor drivers, and regulated power supplies.
• Performed manual soldering, hardware bring-up, and systematic debugging.
• Developed miniature wheeled robots and simple manipulators for experimental validation.

Publications & Research Contributions