finds.dev

reiniscimurs/DRL-robot-navigation

Deep Reinforcement Learning for mobile robot navigation in ROS Gazebo simulator. Using Twin Delayed Deep Deterministic Policy Gradient (TD3) neural network, a robot learns to navigate to a random goal point in a simulated environment while avoiding obstacles.

View on GitHub

★ 1,378PythonMITupdated Dec 2025

// the take

A TD3 (twin delayed DDPG) reinforcement learning implementation that trains a simulated mobile robot in ROS Gazebo to navigate to random goals using laser/Velodyne obstacle data, backed by an ICRA 2022 / IEEE RA-L paper. It's for robotics researchers or grad students who want a working RL navigation baseline to extend, not a drop-in product for a real robot.

Backed by a peer-reviewed paper with the exact same codebase, so the results are at least reproducible rather than a weekend hack claiming SOTA. Uses a 3D Velodyne-style simulated sensor instead of a flat 2D laser, which is closer to what you'd actually deploy on a real platform. Ships a Dockerfile for headless training (no Gazebo/rviz GUI needed), which matters if you want to run this on a remote GPU box or in CI. Tensorboard is wired in from the start so you can actually watch the reward curves instead of guessing.

Hard-pinned to ROS Noetic on Ubuntu 20.04 with Python 3.8 and PyTorch 1.10 — ROS Noetic itself went EOL in May 2025, so you're building on an already-unsupported OS/ROS combo with no ROS2 port in sight. Everything is simulation-only in Gazebo; there's no sim-to-real pipeline, domain randomization, or guidance for deploying the trained policy on actual hardware, so the gap from 'trained in Gazebo' to 'works on a robot' is entirely on you. Process management is a `killall -9` against process names (rosout, gzserver, python, etc.) — fine for a single researcher's machine, a liability anywhere else. TD3 itself is also sample-inefficient and sensitive to reward shaping, so expect to spend real time retuning for any environment that isn't the one in the repo's demo world.

We read the README, the directory tree and a key source file in November 2026. Nobody pays to be listed. How we judge →

// at a glance

Stars
1,378
Forks
195
Language
Python
License
MIT
Last commit
13 Dec 2025

// nearby

If DRL-robot-navigation isn't quite it.

All repos →