Robotics & autonomyProject record · Sep 2026

Robotics & Autonomous Driving

Autonomous driving lab

A ROS2/Gazebo course project connecting lane segmentation, curvature-aware speed control, and LiDAR/GPS localization.

PythonROS2GazeboPyTorchSimpleENetParticle filtersLiDARVehicle control
Why it sits here. The broadest autonomy work: perception, control, localization, and explicit evaluation in one course sequence.
01 / An open road. A closed loop.STUDY IN SPACE

01 / IMPLEMENTATION & CONTRIBUTION

What the work involves

Donald implemented lane-segmentation stages, curvature-based speed control, particle-filter updates, eight-direction LiDAR support, and an offline localization parameter sweep within shared course infrastructure.

Technical depth

Lane masks transformed into bird-eye geometry and fitted lane curves; waypoint lookahead and turn anticipation for speed selection; particle likelihoods with finite-weight safeguards, systematic resampling and GPS reseeding under low effective sample size; eight-direction LiDAR support; offline generated motion/noise trials measuring position RMSE, heading RMSE, convergence and ESS.

The project family

uiuc/ece484

02 / RESULTS

What came out of it

An integrated set of perception, control, and localization modules, with training checkpoints and simulation experiments that compare position error, heading error, convergence, and effective sample size.

03 / SUPPORTING EVIDENCE

Follow the source

Implementation notes, project records, and supporting artifacts.

Source context & project scope

Shared course/team workspace; do not attribute the simulator, model backbone or all source to Donald.

Configured rates and speeds are not measured physical performance.

Offline tuning uses synthetic sequences and cannot substitute for physical vehicle evaluation.

MP0 contains a preliminary decision-rule study with remaining scaffolding; it is not a separately verified autonomous system.

Donald commit 7d45ffc modifies lane-segmentation steps; current node combines segmentation, BEV geometry and fitted lane output.

SOURCE · 2026-09-17

Donald commit c31b8d0 adds curvature-aware longitudinal control; later candidate commits refine the course controller.

SOURCE · 2026-09-17

Donald commit 2f1b315 implements likelihood, normalization, resampling and GPS-related logic.

SOURCE · 2026-09-17

Donald commit f39ba27 adds standalone generated-sequence parameter sweeps and explicit evaluation metrics.

SOURCE · 2026-09-17

Extended LiDAR measurement processing in the attributed localization commit.

SOURCE · 2026-09-17

Preliminary Dubins-mode decision logic with scaffold boundaries.

SOURCE · 2026-09-17
CONTINUE IN ROBOTICS & AUTONOMOUS DRIVING

Reaction Wheel Pendulum