🤖  I teach robots to learn from their mistakes — just like the rest of us.

  About Me

I live in the messy middle ground between simulation and reality — training policies that work beautifully in sim, then spending weeks debugging why they fall apart on real hardware. That's where physics, control theory, and ML collide. That's where I do my best thinking. Grew up in Bangalore. Studied electronics. Realized I wanted to build things that move.

🎓 M.S. Robotics & Autonomous Systems — ASU (3.78) 🔬 RISE Lab — Dr. Wenlong Zhang 💼 Ex-ML Engineer — Jupiter AI Labs 🏅 B.E. ECE — VTU, Bangalore 🎯 Graduating May 2026

  What Drives Me

🧠 Learning Robots Teaching robots to learn from experience through RL — not just follow scripts

🌉 Sim-to-Real Bridging the gap with domain randomization & transfer learning

🦾 Soft Robotics Pneumatic systems for safer, more adaptive human-robot interaction

⚙️ Systems That Ship Production ML that works at scale — not just on a notebook

🔭   What I'm Working On

🔬 Thesis — PRC Physical Reservoir Computing for Soft Robots Idea Use passive pneumatic dynamics as computation Pipeline ROS + ZMQ (100 Hz pressure, 120 Hz mocap) Data 10,000+ synchronized samples Result R² improved 0.80 → 0.90 Target 📄 MECC 2026 ROS ZMQ NumPy SciPy MATLAB

🐕 Quadruped RL Sim-to-Real Transfer for Legged Locomotion Algorithm PPO — 10M iterations Environments 16 parallel sims Success Rate 89% on 5m tasks Hardware 40% less instability Dashboard Joint + LiDAR @ 20 Hz MuJoCo Isaac Sim PyTorch ROS 2

✋ Gesture Arm Hand Tracking → Robot Manipulation Input Hand gesture recognition Output Robotic arm control Goal Intuitive HRI Status 🔨 In development Philosophy Best interface = your hand OpenCV Python ROS Embedded

  Things I've Built

🤖 Fusion Robotics Quadruped + UR5 Mobile Manipulation 🦿 DOF 12 (quad) + 6 (arm) = 18 📡 Comms DDS middleware 🖥️ Monitor Flask + WebSockets ⚡ Setup 45 min → 8 min ROS 2 DDS Flask WebSockets	🐢 Nav-Fusion TurtleBot4 Autonomous Navigation 📍 Position 3cm accuracy ✅ Nav 95% success rate 🗺️ SLAM 5cm grid resolution 📡 Sensors LiDAR + Depth + IMU → EKF ROS 2 Nav2 Cartographer EKF
🔭 Semantic Visual Odometry Drift-Reduced Neural Navigation 📊 Position Error ↓ 34% reduction 🎯 IoU (textureless) 0.78 vs 0.51 baseline 🧠 Architecture OCC-VO + UC-NeRF PyTorch NeRF Visual Odometry CUDA	🎵 VocaLift 🏆 Best Hack for Social Good — DevHacks ASU 🎤 Pitch Real-time extraction 🔊 Audio Meta Audiocraft 🧠 Model LSTM on Mel-spectrograms ♿ Purpose Assistive: text ↔ hum LSTM Audiocraft librosa Flask

🛰️   More — Jupiter AI Labs Production Systems (click to expand)

Aerial Lawn Measurement 🎯 Detection 87% mAP ⚡ Response 450ms 📈 Volume 1,000+ req/day 🏗️ Data 5,000+ aerial images YOLOv8 FastAPI Docker AWS

AI Healthcare Chatbot ✨ Error Reduction 68% 🎯 Intent Accuracy 92% 📚 Documents 50,000 indexed 🔒 Architecture Guardrailed RAG RASA GPT-3.5 ChromaDB RAG

Production infra: AWS · 99.7% uptime · 15,000+ users · CI/CD via GitHub Actions · Prometheus + Grafana

📂   View All Repositories →

🛠️   Tech Stack

🦾   Robotics & Control Frameworks ROS / ROS 2, ZMQ, DDS Simulation MuJoCo, Gazebo, Isaac Sim SLAM & Nav Cartographer, Nav2, ORB-SLAM Sensor Fusion EKF, UKF, Kalman Filters Hardware STM32, ESP32, ADS1115 Manipulators UR5, Custom 12-DOF Quad	🧠   AI / Machine Learning Deep Learning PyTorch, TensorFlow, CUDA RL PPO, SAC, TD3, GAE Computer Vision YOLOv8, Detectron2, OpenCV 3D Vision NeRF, Visual Odometry NLP / LLM Transformers, RAG, LangChain Time-Series LSTMs, AutoTS, ACF
☁️   Cloud & DevOps AWS EC2, S3, SageMaker, Lambda GCP Cloud Run, Vertex AI Containers Docker, Kubernetes CI/CD GitHub Actions, Ansible Monitoring Prometheus, Grafana	💻   Languages & Tools Primary Python, C++, C Web FastAPI, Flask, React Databases PostgreSQL, MongoDB, Redis Scripting Bash, MATLAB Protocols DDS, ZMQ, WebSocket

Python and C++ are my daily drivers. Everything else is a tool I picked up when the problem demanded it.

📊   GitHub Activity

  

🎯   What's Next

🔍 Interested In Sim-to-Real Transfer · Soft Robotics · Multi-Agent Systems · RL for Control

💡 Open To Full-time roles · Research positions · Open-source robotics · Interesting RL problems

📫 Reach Me jeevanhm308@gmail.com jeevan-hm.vercel.app linkedin.com/in/jeevanhm

     

⭐ From Jeevan-HM — Making robots smarter, one gradient descent at a time.