Capabilities
Capability to evidence
Each capability below points to the project that demonstrates it most directly. The detailed pages carry the measurements and implementation details.
Capability Matrix
| Capability | What Shows It | Main Evidence |
|---|---|---|
| Mathematical modeling | Physical dynamics, coordinate residuals, convergence paths, and explicit numerical diagnostics | ballistic-solver |
| API and deployability | Modern C++ API, C ABI, Python/PyPI, Unity/C#, Godot, and ARM64 hardware integration | ballistic-solver |
| ROS 2 / vehicle autonomy integration | FAST-LIO bridge, occupancy mapping, planner integration, measured-delay follower, and Ioniq CAN path | Mapless Autonomous Parking |
| Visual perception | Camera-only metric odometry from ground-plane geometry, dense plane-sweep occupancy, CARLA and 5.2 km real-vehicle evaluation with held-out drives | Monoscale |
| Operational robustness | NTRIP handling, serial retry, timeout detection, live monitoring, and remote deployment workflow | Racing Telemetry Stack |
| Data analysis tooling | Multi-format logs, gate-aligned timing, simulator registration, synchronized media/replay, and report export | Racing Analyze GUI |
| Open-source maintainership | Package maintainership, merged upstream changes, measured integration fixes, and technical code reviews | Autoware autoware_carla_interface maintainer |
Working Principles
- Model the actual bottleneck before choosing the implementation path.
- Make runtime state inspectable through explicit status, logs, plots, replay, or monitoring.
- Package useful cores behind stable APIs, launch paths, services, or analysis workflows.
Evidence Map
- Publication: sole-author ICROS 2026 paper on the solver's residual method, with 0.00% vs. 50.25% failure against direct line-of-sight correction.
- Measured behavior: current solver benchmarks, visual-odometry ablations, telemetry regression tests, research comparisons, and camera-payload measurements.
- Physical validation: real-vehicle parking, a Rock 5B + STM32 solver integration with live vision input, and Monoscale on a 5.2 km real-vehicle dataset.
- Upstream validation: maintainer of Autoware Universe
autoware_carla_interface, eight merged changes, and two technical reviews. - Operational tooling: logs, monitoring, replay, and analysis interfaces that make field behavior inspectable.