LingBot-VLA

AI model
Humanoids and robot learning

LingBot-VLA 2.0 is a Vision-Language-Action foundation model for real-world robot manipulation, pre-trained on about 60,000 hours of robot and egocentric human data.

Made by Robbyant.

Source on GitHub: github.com/Robbyant/lingbot-vla-v2 (opens in a new tab). Facts read at commit be969b8 (opens in a new tab).

Licenses

Code: Apache-2.0 (opens in a new tab)
Read by Robotheca from LICENSE (opens in a new tab) (Apache-2.0) at commit be969b8 (opens in a new tab), on 16 Oct 2026.
Weights: Apache-2.0 (opens in a new tab)
Read on huggingface.co/robbyant/lingbot-vla-v2-6b (opens in a new tab).
Commercial use: Yes

What it does

Pre-trained on about 60,000 hours of data, including 50,000 hours of robot trajectories across 20 robot configurations and 10,000 hours of egocentric human videos.
Its unified 55-dimensional action space covers arms, end-effectors, grippers, dexterous hands, waist, head and mobile-base signals.
Uses a sparse mixture-of-experts action expert to share priors across embodiments and tasks.
Distills current and future perceptual queries from LingBot-Depth and DINO-Video to model scene geometry and evolution.
Post-training example on RoboTwin 2.0 reaches 93.52% clean and 92.80% randomized success with Muon.
Real-robot inference takes about 130 ms per call with 10 denoising steps on an RTX 4090D.

Written by Robotheca from the project's README (opens in a new tab) at commit be969b8.

Getting started

Difficulty: Advanced
Hardware it needs, as the project states it: Requires Python 3.12, PyTorch 2.8.0 and flash-attn 2.8.3 installed through Conda.
How to start: The project's README (opens in a new tab)

Maintenance

Status: Active: a commit in the last 90 days
Newest commit: 22 Sept 2026 on main
Latest release: None published on GitHub

Checked on GitHub on 16 Oct 2026.

Robots on Robotheca that use it

No robot on Robotheca is linked to it yet. Links come from a robot's robot.yaml, from its dependency files read as text, or from Robotheca's editors.

Robotheca lists AI models as a reference: nothing is hosted or run here, and every fact links to where it was read.