Skip to content
RobothecaBeta
Log in

MiMo-Embodied

  • AI model
  • Humanoids and robot learning

An official evaluation suite for the MiMo-Embodied vision-language model, covering embodied AI and autonomous driving benchmarks. It is for evaluation only and contains no training code.

Made by Xiaomi MiMo.

Source on GitHub: github.com/XiaomiMiMo/MiMo-Embodied (opens in a new tab). Facts read at commit 0bee4b6 (opens in a new tab).

Licenses

Code
MIT (opens in a new tab)

Read by Robotheca from LICENSE (opens in a new tab) (MIT) at commit 0bee4b6 (opens in a new tab), on 16 Oct 2026.

Commercial use
Yes

What it does

  • Extends the lmms-eval framework with a custom MiMo model wrapper, mivllm, built on vLLM, for faster data loading and finer image and video preprocessing control.
  • Runs embodied AI benchmarks for affordance prediction, task planning and spatial understanding, such as Where2Place, EgoPlan2, RoboVQA, CV-Bench and VSI-Bench.
  • Runs autonomous driving benchmarks for environmental perception, status prediction, driving planning and driving knowledge QA, such as CODA-LM, DriveLM, nuScenes-QA and LingoQA.
  • Supports single-GPU, multi-GPU and multi-node evaluation, plus batch runs over several tasks.
  • Supports thinking and no-think evaluation modes, with no-think used for some driving and embodied tasks.
  • Provides a launcher script that takes a model path, task name and output directory, and a batch runner that executes tasks sequentially.

Written by Robotheca from the project's README (opens in a new tab) at commit 0bee4b6.

Getting started

Difficulty
Advanced
Hardware it needs, as the project states it
1 × NVIDIA A100 (80GB), or 1 × NVIDIA H20 is recommended; multi-GPU and multi-node setups are supported.

Maintenance

Status
Maintained: a commit in the last 12 months
Newest commit
15 Apr 2026 on main
Latest release
None published on GitHub

Checked on GitHub on 16 Oct 2026.

Links

Robots that can run 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.