Two fronts this week: tidying up the codebase and publishing it as a proper repository, and getting the new target board — the Jetson Nano — past a firmware problem that was stopping Linux from booting. The board is now building and booting.
Until now the code lived as a scatter of scripts on the server. This week I cleaned it up and organised everything into a single, self-contained code repository, separate from this blog — so the actual implementation is easy to read, reuse and review.
The repo is split into three clear layers that mirror the pipeline:
carla-bridge/ — the server-side producers: the two-process CARLA→ROS 2 bridge (raw LiDAR bytes over TCP → PointCloud2) and the earlier odometry bridge.perception_ws/ — the C++ PCL perception node (voxel → RANSAC ground removal → Euclidean clustering → 3D boxes).agl-client/ — the pure-standard-library rosbridge clients that run inside AGL.It ships with a full README (architecture, run instructions, status) and the proof screenshots from Week 3.
Repository: github.com/shauryarane05/agl-carla-perception-GSOC2026 — the complete CARLA → ROS 2 → AGL LiDAR pipeline in one place.
The Jetson Nano ordered last week (after the Raspberry Pi 4B failed to boot) arrived. The first attempts to bring it up didn't go smoothly either: Linux would not boot on the board.
The cause turned out to be outdated firmware on the module — the bootloader/firmware that runs before the OS was too old to bring up the system and load Linux. The fix was to update the firmware first, before flashing and booting the operating system.
Root cause Outdated on-board firmware was preventing Linux from booting. Once the firmware was updated, the board came up normally — the OS itself was never the problem.
With the firmware updated, the Jetson Nano now builds and boots successfully. That clears the hardware blocker that has held up the move off the emulated AGL VM for the last two weeks, and means real-hardware work can finally begin.
Status: Jetson Nano is alive — firmware updated, image building, and the board boots. The target platform is ready for the pipeline.
The goal for the coming week is simple and concrete: replicate the whole pipeline on the Jetson Nano, the same setup proven on the AGL VM, and carry it all the way through — until the board is receiving live LiDAR readings. That means standing up the bridge and rosbridge against the Jetson, and confirming a real point cloud arrives on the hardware target just as it did inside the VM.