Manajev AI Labs

About Me

Based in Zurich, Switzerland.

Hands-on robotics and RL engineer working with ROS 2, C++, and Python for policy deployment, navigation, manipulation, and sim-to-real workflows.

Broad software engineering background across robotics, AI, mobile, web, and blockchain, including many years of full-stack product development.

At Manajev AI Labs, I build reproducible RL training, evaluation, and deployment pipelines, prototype deployable control loops, and work with mobile robots and manipulators in lab environments.

Experience with single- and multi-agent RL, offline/online and model-based methods, and imitation learning in PyTorch and JAX. Open to collaborations, consulting, and full-time roles in Robotics and AI.

Core Expertise

• Reinforcement Learning: PPO, SAC, TD3, MAPPO, MADDPG • Multi-agent (PettingZoo) • Offline/online methods
• Robotics: ROS 2, C++/Python • Nav2, MoveIt 2 • MuJoCo, Isaac Lab • LeRobot • Sim-to-real
• AI/ML: PyTorch, JAX, TensorFlow • Unity ML-Agents • Gymnasium, Weights & Biases
• Infra & DevOps: Docker, CI/CD, AWS • Ansible, Linux
• Product & Leadership: PRDs, roadmaps, Agile • Jira, Confluence
• Languages & Full-Stack: Python, C++, JavaScript/TypeScript, Swift, Solidity • Node.js, React • iOS, Blockchain

Featured Projects

• MARL PPO Suite Clean implementations of PPO-based algorithms (MAPPO, HAPPO) for cooperative multi-agent RL in SMAC environments. Proof: W&B benchmarks, HF Model Zoo.
• Unity Multi-Agent RL Framework Multi-agent reinforcement learning framework implementing MAPPO, MASAC, MATD3, and MADDPG with comprehensive evaluation tools.
• MADDPG Zoo (PyTorch) Modern implementation of MADDPG algorithms using PyTorch and PettingZoo environments for multi-agent RL research.
• RL-Odyssey Research framework for continuous control implementing DDPG, TD3, SAC, PPO, TRPO, A2C, A3C, and PG with interactive notebooks and clean experiment scripts.
• Unity Deep RL (DDPG/TD3/SAC) State-of-the-art deep RL algorithms for continuous control tasks in Unity ML-Agents environments.

• SO-101 ROS Physical AI ROS 2 stack for SO-101 leader/follower arms on real hardware: ros2_control Feetech driver, teleoperation, MoveIt 2, episode recording for imitation learning, and policy inference.
• FastBot Dockerized ROS 2 environment for differential-drive robot: Gazebo sim + real hardware (RPi), Nav2 SLAM, web UI & VPN remote access.
• Manipulation Project ROS 2 manipulation system for UR3E with vision-guided pick-and-place, MoveIt 2 motion planning, and gripper control. Sim + real hardware ready.
• Warehouse Project ROS 2 warehouse automation: AMCL localization, Nav2 path planning, Cartographer SLAM, and custom orchestrator for autonomous shelf pick-and-place.
• ROSbot XL Maze Solver Autonomous maze solver for ROSbot XL using PID control with obstacle avoidance on holonomic base (ROS 2).
• Anthropomorphic Arm IK Inverse Kinematics and motion control of an anthropomorphic 3-DOF manipulator using ROS 1 with Denavit-Hartenberg parameters.

Full code on GitHub. Fork away.

Get in Touch

Email: dmitri@manajev.com

GitHub: @legalaspro

LinkedIn: @dmitri-manajev

Résumé: View online · PDF

Based in Zurich, Worldwide remote.