FMCW LiDAR for Dual-Arm Assembly
GR00T-based peg-in-hole research. The project tests whether range, radial velocity and time data from FMCW LiDAR improve precise insertion by a vision-language-action model.
Who we are looking for
A university lab or a company research team with experience in fine-tuning vision-language-action (VLA) models for embodied AI. We will agree the form of collaboration with each partner.
Project status
This is a proposed research project. The project has no completed model training or robot test results. All performance values below are targets or public reference results.
Research questions
The project will test whether FMCW LiDAR improves dual-arm peg insertion when it supports a pretrained VLA model. It asks three questions:
| Input | Benefit to test |
|---|---|
| Range | Can sparse range measurements improve alignment when stereo depth becomes unreliable? |
| Radial velocity | Can direct speed measurements reduce delay, overshoot or contact force during approach? |
| Time history | Can separate time stamps and motion correction improve coverage without excessive motion blur? |
Radial velocity measures motion along the laser line of sight only. It does not measure full three-dimensional velocity. Sideways motion can produce almost zero radial velocity. A nominal 2 mm range specification does not establish submillimetre assembly accuracy. Force feedback is still necessary inside the hole.
Task and planned setup
Two fixed-base arms pick the parts and move them into a shared view. The left arm holds the hole part. The right arm aligns the peg and inserts it under force control.
| Component | Planned configuration |
|---|---|
| Robot arms | Two Unitree R1-7a arms, each with seven axes and a gripper |
| Camera | Orbbec Gemini 335. Compare passive stereo with active stereo |
| FMCW LiDAR | 100,000 points/s across the full field of view. Non-repeating Lissajous scan, 40° × 40° field |
| Contact and teaching | Two wrist force/torque sensors (pending installation). PICO 4 Ultra controllers for demonstrations |
Model
The main model is GR00T N1.7-3B. The plan freezes the vision-language backbone and trains the full action head. New sensor modules have a combined budget of 30 million parameters. Gated cross-attention lets the action expert read camera and LiDAR features. One policy controls both arms.
Experiments and targets
- Comparison groups: RGB only, passive stereo, active stereo, FMCW range, FMCW full (with radial velocity), and a conventional geometry method.
- Main task: 10 mm peg diameter, 20 mm insertion depth, 0.5 mm diametral clearance. A 0.2 mm clearance tests the capability limit.
- Research target: at least 90% first-attempt success in the main task. This is a target, not a measured result.
- Scale: the core plan contains 3,900 robot trials and about 1,720 accepted demonstration trajectories.
The project brief (PDF) gives the full data plan, training plan and evaluation measures.
Interested in this project?
Send us an email about your team and its relevant work. We will reply and arrange a call to discuss how to work together.
hi@lucidus.tech