RadarSense: LiDAR-driven event reasoning system

RadarSense is a LiDAR-driven event reasoning system that transforms raw spatial motion into high-level interaction events and uses Gemini 3 to infer human engagement states in real time.

Comment

RadarSense

The Story

A curator friend asked me: "I see visitors standing in front of paintings, but I have no idea what they're thinking. Cameras feel invasive. Custom analytics cost too much."

I had a LiDAR sensor lying around — it sees movement as dots, no faces. I thought: what if I just ask Gemini what those movements mean?

I dumped some event data into Gemini 3:

"Person entered slowly, paused 15 seconds, left quickly."

Gemini replied:

"Slow approach suggests curiosity. The pause indicates engagement. Quick exit — either satisfied or uncomfortable."

No custom code. Just reasoning. That's when I knew this could work.

How It Works

LiDAR (dots, no faces) → Event Tokens → Gemini 3 → Insights
Input Output
"Entered, stood 12s, left" "Brief engagement, possibly checking something"
"A waited, B arrived, left together" "Meeting — likely pre-arranged, companions"
"Entered 3 times in 2 min" "Surveillance behavior, anomaly score: high"

The key: New questions = new prompts, not new code.

Why It Matters

Old Way RadarSense
Cameras everywhere LiDAR: no faces, no privacy issues
Weeks of coding per feature Minutes to update a prompt
"Motion detected" "Hesitant approach, then confident exit"

In the Wild: Chengdu Biennale 2026

I built an art installation using RadarSense, now exhibiting at the Chengdu Biennale (Feb 8, 2026 – 6 months, Chengdu, China).

The piece captures visitors' presence and movement in the space, then connects their behavioral "mood" with real-time local weather data. Gemini interprets the interplay — a meditative pause on a rainy afternoon, a hurried crossing during sunny hours — reflecting Chengdu's traditional tea culture philosophy: the harmony between human spirit and the rhythms of heaven and earth.

No complex coding for each weather-mood combination. Just prompts. Gemini handles the poetry.

Applications

Museums, retail, healthcare, security — anywhere you need behavioral insights without surveillance.

Try It

Simulation mode included. No hardware needed. Watch Gemini reason in real-time.

Built With

Share this project:

Updates