Project Report – UWB Indoor Autonomous Navigation

One-paragraph summary

I built an indoor autonomous navigation system that uses dual UWB tags for position, fused odometry for heading, and a custom A* → spline → Pure-Pursuit pipeline to reach user-selected goals on a 2D floorplan. The system communicates with the Iiwari Cloud API to fetch tag positions, integrates with a pre-built AI obstacle detection model over a Nokia 5G link, and visualizes navigation in real time through a lightweight web interface. In live trials, the robot reliably reached its goals with ±0.25 cm accuracy and paused/resumed automatically when obstacles appeared.

Objective

• Evaluate how 10 Hz UWB positioning can be leveraged for indoor autonomous navigation.
• Prove that a simple floorplan-based planner (no LiDAR SLAM) is sufficient for safe autonomy.
• Integrate supervision via an external AI model for pause/resume obstacle handling.
• Demonstrate usability with a live visualizer and real-world demo videos.

What I built (my contributions)

Pose acquisition (via API):

• Queried Iiwari Cloud API for tag status at 10 Hz.
• Front + back UWB tags → midpoint = robot (x,y).
• Heading (yaw) fused from /odometry/filtered (robot_localization).
• Internal transform handled Y-down map vs ROS yaw mismatch.

Planner & controller:

• A* on an occupancy grid from a floorplan PNG (black = obstacle).
• Pruning + Catmull-Rom smoothing for compact, drivable paths.
• Pure-Pursuit controller with:
  • Automatic initial orientation alignment.
  • Curvature-aware dynamic lookahead.
  • Multi-stage docking for precise goal convergence.
• Re-plan when deviation exceeds threshold.

Supervision & control:

• TCP server for pause/resume commands (used by the AI model).
• Safe integration with a Nokia 5G link for reliable, low-latency commands.

Visualization (Flask + Canvas):

• Live map with robot pose, path, and goal.
• Path rendering with glow, curvature thickness, arrows, and trails.
• Telemetry (speed, heading, latency) for operator confidence.

Demonstrations

Normal Navigation (no dynamic obstacles)

• Map included chairs (left) and stairs (right) as obstacles (black areas).
• A* automatically avoided them, producing a safe route.
• Robot smoothly followed the planned path and reached its goal.

Navigation with Obstacle Avoidance

• Same setup, but when a person walked in front of the robot: The external AI (via 5G) paused the robot immediately.
• Once clear, the robot resumed motion and reached the goal.
• Includes a screen recording of the web visualizer showing pose, path, and stop/start events.

Results

• Accuracy: Goal convergence within ±0.25 cm.
• Stability: Smooth control loop despite 10 Hz UWB updates.
• Reliability: Planner respected static obstacles; controller tracked paths tightly.
• Supervision: External AI pause/resume worked seamlessly.
• Usability: Web interface made navigation transparent and operator-friendly.

Why this matters (Impact)

• End-to-end autonomy: sensing (API), planning (A*), control (Pure-Pursuit), visualization, supervision.
• System integration: bridged ROS, cloud APIs, planning, control, 5G, and AI.
• Vendor showcase: Nokia 5G as a robust low-latency backbone.
• HR-ready story: systems thinking, safety, and user-facing tools.

Future Work

• More powerful A*: finer grids/higher compute for smoother paths.
• Improved heading estimation with higher-grade IMU.
• Precision docking for charging/tasks with cm-level repeatability.
• Adaptive obstacle handling: move from pause/resume to dynamic re-planning.

Portfolio one-liner

Built an indoor autonomy stack that uses UWB-based positioning (via cloud API), a custom A*→spline→Pure-Pursuit planner, and Nokia 5G AI integration for obstacle handling — achieving ±0.25 cm precision and delivering a live web-based visualization.