MIKASA-Robo-VLA Documentation
==============================
.. raw:: html
.. raw:: html
Quick Links: :doc:`installation` · :doc:`quickstart` · :doc:`benchmarking` · :doc:`datasets` · `Cite <#citation>`_
MIKASA-Robo-VLA significantly extends `MIKASA-Robo `_ to the VLA setting. It preserves the original benchmark’s focus on memory-intensive tabletop manipulation, while broadening the task suite, introducing language-conditioned evaluation, and providing standardized data export for modern VLA training pipelines.
What changed from MIKASA-Robo (RL release)
- Task set grows from **32 → 90** registered environments covering 10
memory types (vs 4 in the RL release).
- Every task ships a natural-language ``LANGUAGE_INSTRUCTION`` for VLA
conditioning.
- Episodes are grouped into three **horizon splits** (Short / Medium /
Long) so multi-task training and evaluation are tractable.
- 22,500 PPO / motion-planning oracle trajectories are released on
Hugging Face in RLDS and LeRobotDataset v3 formats — no further
conversion needed (6+ millions of transitions).
- Dense and normalised-dense rewards are calibrated for every task,
enabling both offline imitation learning and online RL.
- The original 32-task RL implementation is available from the
`mikasa-robo-rl branch `_
and remains under ``mikasa_robo_suite/rl/`` for backwards compatibility.
Pick your path
--------------
- *"I want to evaluate my VLA model"* → :doc:`benchmarking` (CLI, JSON
output, Python API) and the canonical
:doc:`evaluation_protocol`.
- *"I want to fine-tune a VLA model"* → :doc:`datasets` (RLDS,
LeRobotDataset v3) and :doc:`observation_space`.
- *"I want to explore tasks"* → :doc:`vla_environments/index` (per-task
pages with previews, language instuctions, horizons, and setup parameters).
- *"I want to know what makes the benchmark important"* →
:doc:`concepts` (memory taxonomy, episode structure).
Key Features
------------
- **90 memory tasks** across 10 memory types, horizons 25–2160 steps, multiple difficulty levels.
- The public benchmark grows from 32 (RL release) to **90 tasks** with language instructions for every task.
- Three horizon **splits** (Short / Medium / Long) for structured multi-task evaluation.
- Trajectory collection via PPO oracles and motion planning.
- **22,500 trajectories** (>6 M timesteps) in RLDS and LeRobotDataset v3 formats on Hugging Face.
- Physics fixes, dense / normalised-dense rewards, and full GPU-parallelised simulation via ManiSkill.
.. toctree::
:maxdepth: 2
:caption: Getting Started
installation
quickstart
concepts
.. toctree::
:maxdepth: 2
:caption: The Benchmark
vla_environments/index
benchmarking
evaluation_protocol
observation_space
datasets
.. toctree::
:maxdepth: 2
:caption: Guides
wrappers_cookbook
.. toctree::
:maxdepth: 2
:caption: Reference
api/wrappers
api/collectors
api/envs
faq
Citation
--------
If you use MIKASA-Robo-VLA in your research, please cite:
.. code-block:: bibtex
@inproceedings{cherepanov2026memory,
title = {Memory, Benchmark \& Robots: A Benchmark for Solving Complex Tasks with Reinforcement Learning},
author = {Egor Cherepanov and Nikita Kachaev and Alexey Kovalev and Aleksandr Panov},
booktitle = {The Fourteenth International Conference on Learning Representations},
year = {2026},
url = {https://openreview.net/forum?id=9cLPurIZMj}
}
Legacy RL Version
-----------------
.. note::
If you need the original RL benchmark from the MIKASA-Robo paper
(`arXiv:2502.10550 `_), install
``mikasa-robo-suite==0.0.5`` from PyPI or use the
`mikasa-robo-rl branch `_.
New development targets MIKASA-Robo-VLA. The previous 32-environment RL
implementation is still kept under ``mikasa_robo_suite/rl/`` for
compatibility.
