Internship Progress Weeks 8 & 9 (August 20 ~ September 02)
Following an approved leave in Week 8, this combined report covers the progress made across Weeks 8 and 9. Our primary milestone was upgrading and standardizing the JdeRobot RoboticsAcademy navigation challenges using the ROS 2 Humble Nav2 (Navigation 2) Stack.
We completed the full multi-layered navigation architecture for the City Navigation exercise, validated the pipeline with lead mentor Jose Sir (jmplaza), and started porting the Nav2 stack to the Amazon Warehouse exercise.
1. Video Demonstrations & Previews
A. Initial Overview (City Map & Global Route Planning)
B. Comprehensive Multi-Layered Nav2 Feature Walkthrough
2. City Navigation: Full Nav2 Multi-Layered Architecture
A. Global Path Planning (Navfn A*)
Configured NavfnPlanner on the official JdeRobot City Map ($500\text{ m} \times 500\text{ m}$, resolution $1.25\text{ m/px}$) to compute collision-free, optimal global paths through complex urban road corridors.
B. Dynamic Rolling Local Costmap
Implemented a vehicle-centric rolling local costmap ($35\text{ m} \times 35\text{ m}$) that continuously recalculates obstacle cost gradients and inflation layers around road curbs and building perimeters.
C. Vehicle Geometry & Sensor Integration
- 3D Yellow Taxi URDF: Rendered an authentic $12\text{ m} \times 5\text{ m}$ Yellow Taxi model with dark glass cabin, roof sign, headlights, and 4 black wheels.
- Collision Footprint: Configured an Orange collision bounding polygon matching the taxi chassis dimensions.
- 360-Degree LiDAR Raycasting: Real-time range sensor reflections detecting surrounding road boundaries.
D. Local Trajectory Controller (Regulated Pure Pursuit)
Generated dynamic lookahead steering trajectories (Cyan path arc) corresponding to the local controller output, ensuring smooth cornering at tight intersections.
E. Interactive Goal-Driven Navigation
The vehicle remains safely parked at its initial pose until an interactive destination is dispatched via 2D Goal Pose (/goal_pose) in RViz2. It traverses the planned route and executes an automatic full stop upon arrival at the destination tolerance.
3. Community Review & Mentor Validation
- Presented the complete technical context and video demonstrations on the
#roboticsacademySlack channel. - Clarified implementation details regarding offline
RoboticsInfrastructurelauncher compatibility, layered costmap generation, and local trajectory lookahead curves with lead mentorjmplaza.
4. Amazon Warehouse: Nav2 Integration (In Progress)
Work is currently underway to port and configure the Nav2 navigation stack for the Amazon Warehouse exercise with the Kiva AGV platform.
5. Technical Challenges & Solutions
| Challenge | Root Cause | Solution |
|---|---|---|
| Visual Flickering in RViz2 | Legacy background simulation processes and duplicate RViz instances were publishing conflicting transforms on /tf and /scan. | Terminated all orphan processes, implemented a unified single-process architecture, and applied 3D vertical layer stacking ($Z$-offsets) to prevent GPU depth-buffer collision (Z-fighting). |
| LaserScan Beam Dropping | Default decay time ($0\text{ s}$) caused intermittent frame-rate dropouts over remote display sessions. | Added a $0.2\text{ s} - 0.3\text{ s}$ decay buffer to maintain continuous, solid LiDAR beam visualization. |
6. Next Steps
- Amazon Warehouse Nav2 Integration: Complete and test the Nav2 pipeline for the Amazon Warehouse exercise.
- RoboticsAcademy Package Integration: Standardize launch scripts and configurations for upstream
RoboticsInfrastructure. - Web-GUI Bridge Testing: Verify end-to-end communication with the RoboticsAcademy browser frontend interface via
/webgui/current_target.