# RoboCasa365

RoboCasa365 is a large-scale simulation framework for training and benchmarking **generalist robots** in everyday kitchen tasks. It provides 365 diverse manipulation tasks across 2,500 kitchen environments, with over 3,200 object assets and 600+ hours of human demonstration data. The benchmark tests whether robots can handle the diversity and complexity of real-world household environments.

- Paper: [RoboCasa365: A Large-Scale Simulation Framework for Training and Benchmarking Generalist Robots](https://arxiv.org/abs/2603.04356)
- GitHub: [robocasa/robocasa](https://github.com/robocasa/robocasa)
- Project website: [robocasa.ai](https://robocasa.ai)

## Available tasks

RoboCasa365 includes **365 tasks** organized into atomic (single-skill) and composite (multi-step) categories:

| Category  | Tasks | Description                                                                     |
| --------- | ----- | ------------------------------------------------------------------------------- |
| Atomic    | ~65   | Single-skill tasks: pick-and-place, door/drawer manipulation, appliance control |
| Composite | ~300  | Multi-step tasks across 60+ categories: cooking, cleaning, organizing, etc.     |

**Atomic task examples:** `CloseFridge`, `OpenBlenderLid`, `PickPlaceCoffee`, `ManipulateStoveKnob`, `NavigateKitchen`, `TurnOnToaster`

**Composite task categories:** baking, boiling, brewing, chopping food, clearing table, defrosting food, loading dishwasher, making tea, microwaving food, washing dishes, and many more.

`--env.task` accepts three forms:

- a single task name (e.g. `CloseFridge`)
- a comma-separated list of task names (e.g. `CloseFridge,OpenBlenderLid`)
- a benchmark-group shortcut — `atomic_seen`, `composite_seen`, `composite_unseen`, `pretrain50`, `pretrain100`, `pretrain200`, `pretrain300` — which auto-expands to the upstream task list and auto-sets the dataset `split` (`target` or `pretrain`).

## Installation

RoboCasa and its dependency `robosuite` are not published on PyPI, and RoboCasa's own `setup.py` hardcodes `lerobot==0.3.3`, which conflicts with this repo's `lerobot`. Install them manually as editable clones (using `--no-deps` on `robocasa` to skip its shadowed `lerobot` pin):

```bash
# After following the standard LeRobot installation instructions.

git clone https://github.com/robocasa/robocasa.git ~/robocasa
git clone https://github.com/ARISE-Initiative/robosuite.git ~/robosuite
pip install -e ~/robocasa --no-deps
pip install -e ~/robosuite

# Robocasa's runtime deps (the ones its setup.py would have pulled,
# minus the bad lerobot pin).
pip install numpy numba scipy mujoco pygame Pillow opencv-python \
            pyyaml pynput tqdm termcolor imageio h5py lxml hidapi \
            tianshou gymnasium

python -m robocasa.scripts.setup_macros
# Lightweight assets only (lightwheel registry, ~2GB + the generative
# textures needed by the fixture XMLs). Enough for the default env out
# of the box.
python -m robocasa.scripts.download_kitchen_assets \
  --type tex tex_generative fixtures_lw objs_lw
# Optional: full objaverse/aigen registries (~30GB) for richer object
# variety. If you download these, pass them via `--env.obj_registries`
# at eval/train time (see below).
# python -m robocasa.scripts.download_kitchen_assets --type objs_objaverse
```

<Tip>
RoboCasa365 requires MuJoCo for simulation. Set the rendering backend before training or evaluation:

```bash
export MUJOCO_GL=egl  # for headless servers (HPC, cloud)
```

</Tip>

By default the env samples objects only from the `lightwheel` registry (the one shipped by `--type objs_lw`), which avoids a `Probabilities contain NaN` crash when the objaverse/aigen packs are absent. If you've downloaded the full asset set, enable it with:

```bash
--env.obj_registries='[objaverse,lightwheel]'
```

## Policy inputs and outputs

**Observations** (raw RoboCasa camera names are preserved verbatim):

- `observation.state` -- 16-dim proprioceptive state (base position, base quaternion, relative end-effector position, relative end-effector quaternion, gripper qpos)
- `observation.images.robot0_agentview_left` -- left agent view, 256x256 HWC uint8
- `observation.images.robot0_eye_in_hand` -- wrist camera view, 256x256 HWC uint8
- `observation.images.robot0_agentview_right` -- right agent view, 256x256 HWC uint8

**Actions:**

- Continuous control in `Box(-1, 1, shape=(12,))` -- base motion (4D) + control mode (1D) + end-effector position (3D) + end-effector rotation (3D) + gripper (1D)

## Training on a single task

A ready-to-use single-task dataset is available on the Hub:
[`pepijn223/robocasa_CloseFridge`](https://huggingface.co/datasets/pepijn223/robocasa_CloseFridge).

Train a SmolVLA policy on `CloseFridge`:

```bash
lerobot-train \
  --policy.type=smolvla \
  --policy.repo_id=${HF_USER}/smolvla_robocasa_CloseFridge \
  --policy.load_vlm_weights=true \
  --policy.push_to_hub=true \
  --dataset.repo_id=pepijn223/robocasa_CloseFridge \
  --env.type=robocasa \
  --env.task=CloseFridge \
  --output_dir=./outputs/smolvla_robocasa_CloseFridge \
  --steps=100000 \
  --batch_size=4 \
  --eval_freq=5000 \
  --eval.batch_size=1 \
  --eval.n_episodes=5 \
  --save_freq=10000
```

Evaluate the trained checkpoint on the same task:

```bash
lerobot-eval \
  --policy.path=${HF_USER}/smolvla_robocasa_CloseFridge \
  --env.type=robocasa \
  --env.task=CloseFridge \
  --eval.batch_size=1 \
  --eval.n_episodes=20
```

## Multi-task evaluation

Evaluate across several tasks at once:

```bash
lerobot-eval \
  --policy.path=your-policy-id \
  --env.type=robocasa \
  --env.task=CloseFridge,OpenBlenderLid,PickPlaceCoffee \
  --eval.batch_size=1 \
  --eval.n_episodes=20
```

Or run a full benchmark group:

```bash
lerobot-eval \
  --policy.path=your-policy-id \
  --env.type=robocasa \
  --env.task=atomic_seen \
  --eval.batch_size=1 \
  --eval.n_episodes=20
```

- `--eval.batch_size` controls how many environments run in parallel.
- `--eval.n_episodes` sets how many episodes to run per task.

### Recommended evaluation episodes

For reproducible benchmarking, use **20 episodes per task**. This matches the protocol used in published results.