identity:
name: "Palash Siddharth Mendhe"
role: "Pre-final Year B.Tech in AI & ML @ BIT Mesra (2024β2028)"
leadership: "Vision Team Lead @ Robolution / Pratyunmis (BIT Mesra)"
location: "Ranchi, Jharkhand, India"
timezone: "UTC+05:30"
specializations:
- "Autonomous Mobile Robots (AMR) & Multi-Agent Logistics"
- "Physics-based Simulation (MuJoCo, Gazebo Harmonic, ODE)"
- "Deep RL for Legged Locomotion (CPG + Residual Actor-Critic)"
- "Manipulator Kinematics & Motion Planning (MoveIt 2, Analytical IK)"
loves:
- "Hunting down ODE pivot stalls and tweaking anisotropic friction vectors"
- "Residual Deep RL policies dancing on non-linear Hopf oscillators"
- "Deterministic closed-form IK > numerical solver timeouts"
- "Full-stack robotics CI/CD with Docker & headless testing"
current_vibe: "Listening to Synthwave & tuning reward weights at 2 AM"| Project | Highlights & Tech Stack | Status / Benchmarks |
|---|---|---|
|
π Hybrid CPG-RL Quadruped Locomotion (Unitree Go1) |
β’ Coupled non-linear Hopf/Kuramoto CPG oscillators with residual Deep RL (SAC, TD3, PPO) in MuJoCo for bio-inspired 12-DoF trotting. β’ Benchmarked custom PyTorch algorithms from scratch; multi-objective reward shaping (velocity tracking, clearance symmetry, posture). β’ Tech: PyTorch MuJoCo Gymnasium CPG SAC TD3 PPO CI/CD
|
[ββββββββββ] Completedβ‘ SAC: β₯ 2000 return in 124k steps (~4.5Γ PPO efficiency) β±οΈ TD3: 11.9 ms/step wall-clock π§ͺ 29 Pytest suites + GitHub Actions CI |
|
πͺ Autonomous Mobile Robot & Warehouse Logistics (AMR + Station Manipulators) |
β’ End-to-end fulfillment system coordinating a 4-wheel skid-steer AMR and 6-DOF station arms via asynchronous mission state machine. β’ Diagnosed ODE 89Β° pivot stall: aligned directional friction vectors ( <fdir1>) & decoupled lateral slip (ΞΌβ = 0.05).β’ Rasterized 17 shelf footprints as 100% solid Costmap2D obstacles; monotonic TF broadcaster & +193 mm lift analytical IK. β’ Tech: ROS 2 Jazzy Gazebo Harmonic Nav2 MoveIt 2 AMCL ODE
|
[ββββββββββ] Never Completesπ― 0.02% rotational error with zero drift π¦ Pure-physics parcel retention |
|
π¦Ύ Working perception based locomotion (UR5 Industrial Replica) |
β’ Built a hardened simulation testbed with centralized kinematics, workspace bounds, and motion limits (robot_params.yaml).β’ Implemented closed-form analytical IK ( compute_ik) for deterministic 11-step Cartesian pick-and-place routines.β’ Multi-tier safety architecture: pre-flight action server checks, continuous 3D workspace containment, and automatic emergency abort. β’ Tech: ROS 2 MoveIt 2 Gazebo Harmonic Python Docker Pytest
|
[ββββββββββ] In migration / V3π‘οΈ Continuous 3D safety bounds π³ Automated Docker + GitHub Actions CI |
