Autonomous Mobile Robot System
Building the complete navigation and monitoring infrastructure for a heavy-payload industrial AMR at Tensrai.
Active Commercial Work: Architecture and implementation details are described at a system level — proprietary code and client-specific details are not disclosed.
The Challenge
Industrial AMRs operating in real warehouse environments face problems that standard navigation stacks aren't designed for — dynamic obstacles, narrow corridors, floor surface variations, precise docking requirements, and the need for remote fleet monitoring across facilities. The robot weighs 2500 kg with a heavy payload capacity, which means navigation failures aren't just software bugs — they have real physical consequences.
What I Own — Full Stack Ownership
Navigation Core
Custom Nav2 planner/controller plugin (C++) handling complex warehouse navigation behaviors for heavy industrial AGVs. Standard planners couldn't handle the specific constraints of a 2500kg robot in dynamic industrial environments.
Perception Pipeline
LiDAR-based scan merging and filtering for robust perception in highly dynamic environments. Shadow reduction algorithms to eliminate ghost obstacles from sensor noise. Slope and ramp detection module that dynamically adjusts speed and safety field parameters for terrain-aware navigation.
Localization
Enhanced spatial localization reliability under slippage conditions by integrating raw scan-matching techniques directly with robot odometry. Scan-odometry integration for drift correction in long-run deployments.
Precision Docking
High-accuracy docking strategies using retroreflective tape and LiDAR-based shape recognition for automatic charging and station docking. Millimeter-precision required for reliable autonomous docking.
AI Perception
Integrated YOLO models and Vision-Language Models (VLMs) for real-time object detection, classification, and scene understanding in warehouse environments. Enables context-aware navigation decisions beyond simple obstacle avoidance.
Cloud Dashboard (Full Backend)
Architected and built the complete backend for the robot monitoring dashboard:
- → AWS EC2 deployment and server management
- → WebSocket bridge between ROS2 and cloud (real-time telemetry)
- → IP management and network configuration for multi-machine communication
- → Data flow optimization to minimize latency on live robot state
- → Remote fleet monitoring — operators see robot status, missions, and alerts
From the robot's sensors to the operator's screen — I own the entire pipeline.
Deployment
Led end-to-end testing, hardware-in-the-loop validation, and final real-world site deployments at customer warehouse locations. Currently in active deployment phase.
The Stack
Navigation
ROS2, Nav2, Custom Plugins, BT.CPP, TF2, Lifecycle Nodes
Perception
LiDAR, OpenCV, YOLO, Vision-Language Models
Localization
FAST-LIO, AMCL, Scan-Odometry Fusion
Cloud
AWS EC2, WebSocket, ROS2 Networking, Multi-machine Comms
Programming
C++ (ROS2), Python, Bash
Hardware
2500kg AMR, Ouster LiDAR, Depth Cameras, Retroreflective Docking
Status
Deployment video coming soon
Recording the robot operating in the real warehouse environment post-deployment. Will be added here.