KO | EN

GitLite

Minimalist Open Source Repository Search Engine

gitlite — search

#typescript #ai-agents #developer-tools #cli #codex #claude-code #react #dsh-plugin #mcp #open-source #dsh #deepseek-harness

scaling-crl

★ 309 Open GitHub ↗

No description available.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# 1000 Layer Networks for Self-Supervised RL: Scaling Depth Can Enable New Goal-Reaching Capabilities

> [!IMPORTANT]  
> [Our work was selected for the Best Paper Award at NeurIPS 2025!](https://blog.neurips.cc/2025/11/26/announcing-the-neurips-2025-best-paper-awards/#:~:text=1000%20Layer%20Networks%20for%20Self%2DSupervised%20RL%3A%20Scaling%20Depth%20Can%20Enable%20New%20Goal%2DReaching%20Capabilities) 🥳

Email kw6487@princeton.edu with questions/comments/suggestions.

![Environments](https://raw.githubusercontent.com/wang-kevin3290/scaling-crl/main/assets/envs.gif)
Our work builds on top of [JAXGCRL](https://github.com/MichalBortkiewicz/JaxGCRL), feel free to check it out!!

# Installation

```sh
uv sync
```
Then just fix the two Brax issues described below, and you'll be all set.

## Fixing two bugs in brax 0.10.1
1. There is a minor bug in brax's contact.py file. To fix it, first locate the brax contact.py file in your virtual environment: 
```
find .venv -name contact.py
```
Then open the file and replace it with the following code:
```python
from typing import Optional
from brax import math
from brax.base import Contact
from brax.base import System
from brax.base import Transform
import jax
from jax import numpy as jp
from mujoco import mjx

def get(sys: System, x: Transform) -> Optional[Contact]:
    """Calculates contacts.
    Args:
        sys: system defining the kinematic tree and other properties
        x: link transforms in world frame
    Returns:
        Contact pytree
    """
    #NOTE: THIS WAS MODIFIED SINCE AFTER MUJOCO 3.1.5, mjx.ncon IS NOT AVAILABLE
    # ncon = mjx.ncon(sys)
    # if not ncon:
    #   return None
    data = mjx.make_data(sys)
    if data.ncon == 0:
        return None
    @jax.vmap
    def local_to_global(pos1, quat1, pos2, quat2):
        pos = pos1 + math.rotate(pos2, quat1)
        mat = math.quat_to_3x3(math.quat_mul(quat1, quat2))
        return pos, mat
    x = x.concatenate(Transform.zero((1,)))
    xpos = x.pos[sys.geom_bodyid - 1]
    xquat = x.rot[sys.geom_bodyid - 1]
    geom_xpos, geom_xmat = local_to_global(
        xpos, xquat, sys.geom_pos, sys.geom_quat
    )
    # pytype: disable=wrong-arg-types
    d = data.replace(geom_xpos=geom_xpos, geom_xmat=geom_xmat)
    d = mjx.collision(sys, d)
    # pytype: enable=wrong-arg-types
    c = d.contact
    elasticity = (sys.elasticity[c.geom1] + sys.elasticity[c.geom2]) * 0.5
    body1 = jp.array(sys.geom_bodyid)[c.geom1] - 1
    body2 = jp.array(sys.geom_bodyid)[c.geom2] - 1
    link_idx = (body1, body2)
    return Contact(elasticity=elasticity, link_idx=link_idx, **c.__dict__)
```
2. There is also a minor bug in brax's json.py file. To fix it, first locate the brax json.py file in your virtual environment:
```
find .venv -name json.py | grep "/brax/io/json.py"
```
Then open the file and change the if statement in line 159 to:  
```python
if (rgba == jp.array([0.5, 0.5, 0.5, 1.0])).all():
```

# Running experiments
Now, we are ready to run the train script. To run the code, you'll need a GPU. For Humanoid-based environments, it may require up to 80GB of GPU memory (for deep networks). Below is an example command to run the training script (an additional example can be found in the provided slurm script `job.slurm`):

```sh
uv run train.py --env_id "humanoid" --eval_env_id "humanoid" --num_epochs 100 --total_env_steps 100000000 --critic_depth 16 --actor_depth 16 --actor_skip_connections 4 --critic_skip_connections 4 --batch_size 512 --vis_length 1000 --save_buffer 0 
```

>[!NOTE]
>If you would like the experiments to be synced to wandb, you should go to `train.py` and replace the default values of `wandb_entity` and `wandb_project_name` (line 34-35 of the `train.py` file) with your particular wandb entity and wandb project name. Alternatively, these two can also be set as hyperparameter flags when running the train script.

# Citing Scaling CRL 📜
```bibtex
@inproceedings{wang2025,
  title     = {1000 Layer Networks for Self-Supervised {RL}: Scaling Depth Can Enable New Goal-Reaching Capabilities},
  author    = {Kevin Wang and Ishaan Javali and Micha{\l} Bortkiewicz and Tomasz Trzcinski and Benjamin Eysenbach},
  booktitle = {The Thirty-ninth Annual Conference on Neural Information Processing Systems},
  year      = {2025},
  url       = {https://openreview.net/forum?id=s0JVsx3bx1}
}
```

<!-- 
## Troubleshooting Potential Errors

**If you encounter the following error:**
```AttributeError: module 'mujoco.mjx' has no attribute 'ncon'```

**Fix:**
1. Locate the brax contact.py file in your conda environment: 
   ```
   find ~/.conda/envs/scaling-crl -name contact.py
   ```
2. Open the file and replace it with the following code:

```python
    from typing import Optional
    from brax import math
    from brax.base import Contact
    from brax.base import System
    from brax.base import Transform
    import jax
    from jax import numpy as jp
    from mujoco import mjx

def get(sys: System, x: Transform) -> Optional[Contact]:
        """Calculates contacts.
        Args:
            sys: system defining the kinematic tree and other properties
            x: link transforms in world frame
        Returns:
            Contact pytree
        """
        #NOTE: THIS WAS MODIFIED SINCE AFTER MUJOCO 3.1.5, mjx.ncon IS NOT AVAILABLE
        # ncon = mjx.ncon(sys)
        # if not ncon:
        #   return None
        data = mjx.make_data(sys)
        if data.ncon == 0:
            return None
        @jax.vmap
        def local_to_global(pos1, quat1, pos2, quat2):
            pos = pos1 + math.rotate(pos2, quat1)
            mat = math.quat_to_3x3(math.quat_mul(quat1, quat2))
            return pos, mat
        x = x.concatenate(Transform.zero((1,)))
        xpos = x.pos[sys.geom_bodyid - 1]
        xquat = x.rot[sys.geom_bodyid - 1]
        geom_xpos, geom_xmat = local_to_global(
            xpos, xquat, sys.geom_pos, sys.geom_quat
        )
        # pytype: disable=wrong-arg-types
        d = data.replace(geom_xpos=geom_xpos, geom_xmat=geom_xmat)
        d = mjx.collision(sys, d)
        # pytype: enable=wrong-arg-types
        c = d.contact
        elasticity = (sys.elasticity[c.geom1] + sys.elasticity[c.geom2]) * 0.5
        body1 = jp.array(sys.geom_bodyid)[c.geom1] - 1
        body2 = jp.array(sys.geom_bodyid)[c.geom2] - 1
        link_idx = (body1, body2)
        return Contact(elasticity=elasticity, link_idx=link_idx, **c.__dict__)
    ```
3. Save the file and rerun the training script.

**If you encounter the following error:** ```Error rendering final policy: unsupported operand type(s) for ==: 'ArrayImpl' and 'list'```

**Fix:**
1. Locate the brax json.py file in your conda environment:
   ```
   find ~/.conda/envs/scaling-crl -name json.py | grep "/brax/io/json.py"
   ```
2. Open the file and change the if statement in line 159 to:
    ```python
    if (rgba == jp.array([0.5, 0.5, 0.5, 1.0])).all():
    ```
3. Save the file and rerun the training script. -->

scaling-crl

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README