Internship Progress Week 16 (October 15 ~ October 21)

Nav2 Global Navigation, Follow Line Demonstration on macOS, and Monaco 3D Circuit Benchmark

This report covers the progress made during Week 16 (October 15 ~ October 21, 2026). Our work this week focused on:

1. Validating and recording demonstrations for Nav2 Global Navigation and Follow Line running on macOS (Apple Silicon) without dedicated GPU.
2. Benchmarking the Monaco 3D Circuit performance and identifying key bottlenecks (Issue #3493 / RoboticsInfrastructure #632).

---

1. Video Demonstrations

Nav2 Global Navigation on macOS (Apple Silicon, CPU Runtime)

Demonstration of autonomous navigation using ROS 2 and Navigation2 (Nav2) inside the updated RADI container on Apple Silicon without dedicated GPU acceleration.

Direct YouTube Link: https://youtu.be/ERUeN0VsJEE

Key Highlights:

• Stack: ROS 2 + Nav2 (Navigation2) action client architecture.
• Features: Real-time global costmap, local costmap, and dynamic obstacle avoidance.
• Platform: Gazebo Sim running on macOS via Docker CPU runtime.

---

Follow Line on macOS with ROS 2 Lyrical & Gazebo Jetty

Demonstration of the Follow Line exercise running in ROS 2 Lyrical and Gazebo Jetty on macOS without GPU, validating camera stream capture, real-time image processing, and RTF performance.

Direct YouTube Link: https://youtu.be/TeJxnLOyRio

Key Highlights:

• Stack: ROS 2 Lyrical and Gazebo Jetty.
• Vision: OpenCV HSV color filtering and centroid tracking running at 30+ Hz.
• Performance: Maintained stable Real-Time Factor (RTF ~ 0.95 - 1.05) on macOS CPU.

---

2. Monaco 3D Circuit Benchmark (Issue #3493)

We profiled and benchmarked the Monaco 3D Circuit (Issue #3493 / RoboticsInfrastructure #632):

• Progress Made: Replaced the heavy 31k-triangle Car.obj collision mesh with primitive colliders (box chassis, sphere wheels) in f1_common.urdf.xacro, generated a clean 3D collision mesh (monaco_collision.dae) with 1.2m boundary walls, and calibrated 2D map tracking in showCarPositionFollowLine.js.
• Current Failure Point: The vehicle navigates the first half of the track smoothly (Sainte Devote, Beau Rivage crest, Casino Square), but currently runs off track at the sharp downhill Fairmont Hairpin (180° turn) due to short camera lookahead and line loss at speed. Additionally, live RTF dips to ~0.60 on track due to the 250 Hz physics step size (max_step_size 0.004).
• Planned Fix:
  1. Update physics step size to 0.008 (125 Hz) in monaco_line.world to cut solver load in half and lock RTF at a steady 1.00.
  2. Implement contour-based nearest-line tracking and predictive deceleration to slow the car ahead of hairpins and complete full laps reliably.

---

3. Summary & Next Steps

• Published demonstration videos for Nav2 Global Navigation and Follow Line on macOS.
• Identified physics step size and hairpin steering lookahead as the core bottlenecks for Monaco.
• Implementing 125 Hz physics and contour-based cornering control to complete full Monaco laps.