NavBot-X Isaac Lab is a flagship robotics simulation project built around autonomous rover navigation, reinforcement learning, and human-in-the-loop control inside a premium futuristic industrial environment called Aegis Inspection Bay.
This project is the next-generation continuation of an earlier Gazebo-based navigation system, now rebuilt in an Isaac Sim / Isaac Lab / Omniverse-oriented stack to support better simulation quality, stronger project identity, and a more advanced Physical AI portfolio direction.
The purpose of NavBot-X Isaac Lab is to build a compact autonomous inspection rover that can operate in two control modes:
- Self-controlled mode using Reinforcement Learning
- Human-guided mode using gesture-based control
This dual capability is what makes the project unique.
The rover is designed to perform structured indoor inspection-style navigation tasks across multiple checkpoints in a futuristic industrial environment. It can learn to make movement decisions by itself through PPO-based RL, while also supporting human override / human-directed behavior through gesture input.
What makes NavBot-X Isaac Lab unique is its hybrid intelligence model:
- Autonomous control through Reinforcement Learning
- Human control through gesture-based command input
- Premium simulator-side environment design instead of a bare utility scene
- Checkpoint mission progression rather than only one-point navigation
- Physical AI direction combining perception-inspired design, embodied movement, route logic, and simulation-driven control
This project is not just “a robot moving in simulation.”
It explores a more meaningful robotics idea:
a rover that can act on its own or be guided by a human, depending on the task and control mode.
NavBot-X Isaac Lab is useful as a prototype for:
- industrial inspection robotics
- indoor route monitoring
- autonomous checkpoint traversal
- human-in-the-loop robotic supervision
- simulation-first robotics development
- RL policy learning in structured environments
- future Physical AI systems that combine autonomy and human intervention
This kind of architecture is relevant for environments where a robot should usually work independently, but still allow a person to guide or override it when needed.
Examples:
- factory inspection bays
- warehouse patrol systems
- industrial monitoring zones
- robotics testing environments
- future smart facility agents
This repo represents the progression from an earlier Gazebo foundation into a more advanced Isaac-based robotics simulation stack.
The earlier project established the navigation and RL idea in a Gazebo-based setup.
The first Isaac phase focused on scene building, rover identity, cameras, cinematic route presentation, and premium environment design.
Phase 2 introduced checkpoint progression, task sequencing, and more meaningful inspection mission behavior.
Phase 3 focused on PPO-based RL training for sequential checkpoint traversal.
Alongside autonomous behavior, gesture input was explored as a human-guided control mode.
At the end of this milestone, NavBot-X achieved:
- autonomous sequential checkpoint traversal using PPO
- successful multi-stage route learning in Isaac Sim
- stable mission progression across checkpoints
- dual-control project identity:
- RL-based autonomous mode
- gesture-based human control mode
This project is now being frozen as a completed milestone release before future extensions.
- Ubuntu 24.04
- Isaac Sim
- Isaac Lab
- Python
- PyTorch
- PPO (Proximal Policy Optimization)
- NumPy
- Omniverse / USD scene construction
- Custom rover controller
- Gesture input pipeline (separate control path)
NavBot-X is a compact inspection rover with a modular structure consisting of:
- Main Chassis
- Upper Equipment Deck
- Front Sensor Module
- Central Sensor Mast
- Sensor Head / Perception Pod
- Wheel Drive Modules
- Undercarriage Mobility Frame
- Signature Tech Accent Elements
The rover is visually designed as a premium futuristic inspection platform and functionally structured as a checkpoint-navigation agent.
The reinforcement learning task evolved in stages:
- single checkpoint navigation
- A → B sequential navigation
- A → B → C sequential navigation
The policy learns to map normalized navigation observations to rover actions:
- forward motion
- yaw / turning control
The training loop uses PPO with:
- policy network
- value network
- rollout collection
- GAE / returns
- minibatch PPO updates
- checkpoint saving and training continuation
The rover learns to move toward checkpoints by maximizing reward through trial and error.
The rover can also be controlled through a human command pipeline, giving the project a human-in-the-loop control dimension.
This combination is the strongest identity feature of the project.
navbot-x-isaac-lab/
├── README.md
├── LICENSE
├── .gitignore
├── pyproject.toml
├── setup.py
│
├── assets/
├── config/
├── docs/
├── experiments/
│
├── logs/
│ └── .gitkeep
├── checkpoints/
│ └── .gitkeep
│
├── media/
│ ├── images/
│ │ ├── foundationphase.png
│ │ ├── hero.png
│ │ ├── route.png
│ │ ├── phasetwo.png
│ │ ├── RL.png
│ │ └── gesture.png
│ ├── videos/
│ │ └── .gitkeep
│ ├── renders/
│ │ └── .gitkeep
│ └── demo_clips/
│ └── .gitkeep
│
├── scripts/
│ └── run_aegis_app.py
│
├── tools/
│ ├── gesture_writer.py
│ ├── ppo_train_navbot.py
│ ├── rl_env_sanity_check.py
│ └── rl_train_stub.py
│
├── source/
│ └── navbot_x_isaac_lab/
│ ├── __init__.py
│ ├── utils/
│ │ ├── __init__.py
│ │ └── paths.py
│ │
│ ├── apps/
│ │ ├── __init__.py
│ │ └── aegis_app.py
│ │
│ ├── demo/
│ │ ├── __init__.py
│ │ ├── aegis_scene_builder.py
│ │ ├── checkpoint_manager.py
│ │ ├── gesture_controller.py
│ │ ├── gesture_input.py
│ │ ├── inspection_mission.py
│ │ ├── inspection_task.py
│ │ ├── layout_blueprint.py
│ │ ├── route_animator.py
│ │ ├── rover_controller.py
│ │ ├── scene_config.py
│ │ └── status_visualizer.py
│ │
│ ├── robots/
│ │ ├── __init__.py
│ │ └── mobile_base/
│ │
│ └── tasks/
│ ├── __init__.py
│ └── navigation/
│ ├── __init__.py
│ ├── navigation_env.py
│ ├── navigation_env_cfg.py
│ ├── inspection_rl_env.py
│ ├── inspection_rl_env_cfg.py
│ └── agents/
Copyright (c) 2026 Atharva Sharma. All rights reserved.
This repository is shared for portfolio and demonstration purposes only. No part of this project may be copied, redistributed, modified, or used commercially without explicit written permission from the author.