Sensor Fusion for Distance Estimation (LiDAR + Sonar + Kalman)

Robust short-range distance estimation using LiDAR + ultrasonic sonar fusion, improved via multipoint calibration, noise-based weighting, and a 1-D Kalman filter for real-time smoothing. Includes an RGB distance indicator demo for intuitive obstacle awareness.

Built as a Sensing & Sensor Fusion course project (ARU). Focus: practical reliability under surface-dependent sensing failure modes.

---

Why this matters (engineering problem)

Single sensors fail in predictable ways:

• LiDAR (ToF) can drop out or spike on dark/absorptive or angle-dependent reflective surfaces.
• Sonar is generally colour-independent but can jitter with geometry, edge effects, or multi-path.
• Analogue IR distance is often nonlinear and surface-dependent at short range (not reliable for quantitative fusion in this build).

This repo shows a simple, implementable pipeline that keeps distance estimation usable even when one sensor becomes unreliable.

---

System overview

Pipeline

1. Multipoint calibration (reduce offset/bias)
2. Sensor behaviour profiling (surface + range)
3. Weighted fusion (reliability-based blending)
4. 1-D Kalman filtering (smooth fused estimate + provide uncertainty)
5. RGB indicator demo (Green/Yellow/Red distance zones)

Hardware

• Arduino UNO
• LiDAR ToF sensor (VL53-series class)
• Ultrasonic sonar (HC-SR04 class)
• IR module (used as digital trigger)
• RGB LED module, breadboard, jumper wires, USB

Software

• Arduino IDE (sampling + serial logging + demo logic)
• MATLAB (visualisation, RMSE/analysis, plots)
• Excel (data logging/csv prep)

---

Key features

• ✅ Surface-aware robustness: fusion falls back to the more reliable sensor when the other drops out.
• ✅ Noise reduction: Kalman filter improves stability for downstream logic (e.g., thresholds).
• ✅ Reproducible experiment design: controlled distances and surfaces with consistent sampling.
• ✅ Demo-ready output: RGB indicator based on fused distance zones.

---

Results at a glance (what to look for)

Experiments were run at 5 cm, 15 cm, 30 cm on white cardboard, black cloth, metal.

Typical observations:

• Black cloth: LiDAR often becomes unreliable (low reflectivity) → fusion becomes sonar-dominant; Kalman smooths jitter.
• Metal: both sensors usable but LiDAR can show angle-dependent artefacts → fusion + Kalman converges between sensors.
• White: both sensors strong; filter rejects early outliers/spikes and stabilises quickly.

See docs/results.md and exported plots in results/.

---

---

How to run (end-to-end)

A) Arduino: capture distance streams

1. Wire sensors to the UNO (see hardware/wiring.md).
2. Upload src/arduino/main.ino.
3. Open Serial Monitor / log serial output to CSV.

B) MATLAB: visualise + analyse

1. Put captured CSVs into data/
2. Run:
   • src/matlab/plot_raw_vs_fused.m
   • src/matlab/kalman_filter_1d.m
   • src/matlab/rmse_comparison.m
3. Export plots to results/.

---

Engineering notes (important details)

Weighted fusion (noise-based)

For sensors i ∈ {lidar, sonar}, fused distance: [ \hat{x} = \frac{\sum_i w_i x_i}{\sum_i w_i} ] Where weights reflect reliability (commonly inverse variance): [ w_i = \frac{1}{\sigma_i^2} ]

Why fused RMSE can be higher (a real failure case)

Fusion does not automatically cancel error if sensors share similar bias or if weights are mis-specified for a given surface/range. In that case the fused estimate can “inherit” systematic error from both sensors.

---

Limitations

• Fixed noise-based weights do not fully adapt to changing surfaces/angles.
• Hardware alignment and mounting can introduce repeatable bias.
• Range tested was narrow (5–30 cm); wider ranges and moving targets were not evaluated here.

Roadmap

• Adaptive weighting (surface-aware / innovation-based)
• Wider distance range + more materials + angle sweeps
• Moving-target tests
• Replace analogue IR with digital ToF or use IR strictly as trigger

---

Quick resume bullets (paste into CV)

• Built an embedded distance-estimation pipeline using LiDAR + sonar fusion with multipoint calibration, noise-based weighting, and 1-D Kalman filtering for stable real-time output.
• Designed controlled experiments across three surfaces and three ground-truth distances, logging and analysing sensor behaviour with MATLAB.
• Implemented an RGB distance indicator demo driven by fused distance zones for obstacle awareness.

---

References

• STMicroelectronics VL53-series ToF sensor datasheet (VL53L0X class)
• HC-SR04 ultrasonic ranging module datasheet
• Grewal & Andrews, Kalman Filtering: Theory and Practice Using MATLAB

Hardware build & test evidence

Bench prototype (Arduino + sensors):

Repeatable distance testing using tape measure (ground truth) for black, white and metal:

Presentation of results (evidence)

Photo evidence of presenting the project: