---
title: 3D HAMSTER — 3D Trajectory Prediction
emoji: 🐹
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 6.15.1
app_file: app.py
short_description: Predict metric 3D robot trajectories from RGB-D + language
python_version: "3.12"
startup_duration_timeout: 1h
pinned: false
license: apache-2.0
models:
  - DAVIAN-Robotics/3D_HAMSTER
---

# 🐹 3D HAMSTER — 3D Trajectory Prediction

Interactive demo for **3D HAMSTER**, a depth-aware Vision-Language-Action planner
that predicts **metrically grounded 3D end-effector trajectories** directly from a
single RGB-D observation and a language instruction.

- **Backbone**: Qwen3-VL-8B + a frozen LingBot-Depth (DINOv2 ViT-L/14) geometry encoder.
- **Inputs**: RGB image + metric depth map (`.npy`, meters, aligned to RGB) + instruction.
- **Outputs**: `[u, v, depth]` waypoints + gripper actions, visualized as a 2D overlay
  and an interactive 3D scene + trajectory.
- **Modes**: 3D Trajectory (default), 2D Trajectory, 3D/2D Pointing, 2D Bounding Box, General VQA.

> ⚠️ Depth must be **metric (meters)** and aligned to the RGB frame.

Links: [Paper](https://huggingface.co/papers/2606.31329) ·
[Model](https://huggingface.co/DAVIAN-Robotics/3D_HAMSTER) ·
[Code](https://github.com/DAVIAN-Robotics/3D_HAMSTER) ·
[Project page](https://davian-robotics.github.io/3D_HAMSTER/)