Notes
Short technical writing on systems, perception, benchmarking, and agentic engineering.
Why Multi-Camera Multi-Sensor SLAM Matters
Robust SLAM systems benefit from complementary sensing because single-camera pipelines often fail exactly where real-world deployment becomes most demanding.
Why Maintainability Benchmarks Matter for Coding Agents
Functional correctness alone misses whether an agent’s changes remain local, reusable, testable, and structurally coherent inside a real codebase.
Why Image-Based Gravity Estimation Is Useful for VIO and SLAM
A single image can provide a useful geometric prior when inertial gravity estimates become unreliable under acceleration, vibration, or transient motion.