WALL-OSS
X Square Robot's open embodied foundation model: a Qwen2.5-VL-3B backbone with mixture-of-experts action routing, trained in two stages (discrete FAST action tokens, then flow matching). It ships as two checkpoints, a flow-matching one (wall-oss-flow) and a FAST-token one (wall-oss-fast).
open weightsverified 2026-10-24paper →
Architecture
Compute requirements
| Memory | Precision | Conditions | |
|---|---|---|---|
| train | ≥24 GB · ≤40 GBsrc | — | Peak training VRAM at batch size 8 with AdamW (LeRobot default), "Large VLA" group (`pi0`, `pi0_fast`, `pi05`, `xvla`, `wall_x`) in the LeRobot compute hardware guide; "A100 40 GB+ (24 GB tight at BS 1)". The guide calls its figures indicative, order-of-magnitude numbers for picking hardware, not exact predictions. Precision not stated. · batch 8 |
Training requirements: see also the LeRobot compute hardware guide, the authoritative source for LeRobot training envelopes.
Measured inference 0 runs · inference cost only, not task success
No measured runs yet. Submit one with its runtime, precision and camera count.
Robots running WALL-OSS 4 of 24 with evidence
| Robot | State | Tier | Verified | Detail |
|---|---|---|---|---|
| AgileX PiPER | partial | T3 · community reported | 2026-10-24 1d ago | Training evidenced by public checkpoint(s); on-robot inference not documented by a primary source. |
| ROBOTIS OMX | partial | T3 · community reported | 2026-10-25 today | Training evidenced by public checkpoint(s); on-robot inference not documented by a primary source. |
| SO-100 | partial | T3 · community reported | 2026-10-24 1d ago | Training evidenced by public checkpoint(s); on-robot inference not documented by a primary source. |
| SO-101 | partial | T3 · community reported | 2026-10-24 1d ago | Training evidenced by public checkpoint(s); on-robot inference not documented by a primary source. |
No verified evidence, but both are available in LeRobot (ecosystem context): ARX5, Bimanual SO-100/SO-101, EarthRover Mini Plus, HopeJR, I2RT YAM, Koch v1.1, LeKiwi, OpenArm, Reachy 2, reBot Arm B601-DM, Trossen WidowX AI, Unitree G1, Universal Robots UR5e, UFACTORY xArm, XLeRobot.
Framework support
| Framework | Support | Since | Type string | Docs |
|---|---|---|---|---|
| LeRobot | nativesrc | 0.4.3 | wall_x | docs |
| LeRobot: Policy type `wall_x` (directory src/lerobot/policies/wall_x, extra `.[wallx]`); docs page and README title it WALL-OSS. `--policy.prediction_mode` selects `diffusion` (flow matching, default) or `fast` (next-token prediction). Default chunk_size 32. Only attn_implementation='eager' is accepted in v0.6.1. | ||||
Checkpoints
- x-square-robot/wall-oss-flow · Flow-matching checkpoint; default `pretrained_name_or_path` in the LeRobot wall_x config. No license tag on the Hub.
- x-square-robot/wall-oss-fast · FAST-token checkpoint.
- x-square-robot/wall-oss-flow-0.1
Sources
- T1 · upstream documentedpapersource 2025-09-15verified 2026-10-24 (1d ago)
WALL-OSS uses Qwen2.5-VL-3B as its main backbone.
“We adopt QwenVL2.5-3B as the main backbone.”
https://arxiv.org/abs/2509.11766 - T1 · upstream documentedpapersource 2025-09-15verified 2026-10-24 (1d ago)
The Integration stage swaps discrete action prediction for flow-matching continuous actions.
“we replace discrete action prediction with continuous action modeling via flow matching”
https://arxiv.org/abs/2509.11766 - T1 · upstream documentedofficial-docssource 2026-08-03verified 2026-10-24 (1d ago)lerobot 0.6.1
LeRobot WALL-OSS README lists both action prediction modes.
“| Action Prediction | Flow Matching (diffusion) or FAST (discrete tokens) |”
https://github.com/huggingface/lerobot/blob/v0.6.1/docs/source/policy_walloss_README.md - T1 · upstream documentedofficial-docssource 2026-08-03verified 2026-10-24 (1d ago)lerobot 0.6.1
LeRobot docs describe WALL-OSS as adapted from the WallX repository; license Apache 2.0.
“The LeRobot implementation is adapted from their open-source [WallX](https://github.com/X-Square-Robot/wall-x) repository.”
https://github.com/huggingface/lerobot/blob/v0.6.1/docs/source/walloss.mdx - T1 · upstream documentedsource-codesource 2026-08-03verified 2026-10-24 (1d ago)lerobot 0.6.1
Policy type `wall_x`; requires at least one visual input feature; max_state_dim 20; chunk_size 32.
“"Wall-X policy requires at least one visual input feature. "”
https://github.com/huggingface/lerobot/blob/v0.6.1/src/lerobot/policies/wall_x/configuration_wall_x.py - T1 · upstream documentedsource-codesource 2026-08-03verified 2026-10-24 (1d ago)lerobot 0.6.1
Policy type string is `wall_x`.
“@PreTrainedConfig.register_subclass("wall_x")”
https://github.com/huggingface/lerobot/blob/v0.6.1/src/lerobot/policies/wall_x/configuration_wall_x.py - T1 · upstream documentedrelease-notessource 2026-01-22verified 2026-10-24 (1d ago)lerobot 0.4.3
WALL-OSS (wall_x) was added in LeRobot v0.4.3 (PR
“feat: support wallx model by @wut19”
https://github.com/huggingface/lerobot/releases/tag/v0.4.3 - T1 · upstream documentedofficial-docssource 2026-09-11verified 2026-10-24 (1d ago)
Wall-X is the codebase for the WALL model series and lists the WALL-OSS checkpoints.
“This repository provides the training and inference code for the WALL series”
https://github.com/X-Square-Robot/wall-x/blob/ce2a1ae31d9ee1028b97ca40e3f66034cf8c0e5f/README.md - T1 · upstream documentedofficial-docssource 2026-08-03verified 2026-10-24 (1d ago)lerobot 0.6.1
Compute hardware guide: Large VLA group peak VRAM ~24–40GB at BS 8 with AdamW; figures are indicative.
“| Large VLA | `pi0`, `pi0_fast`, `pi05`, `xvla`, `wall_x` | ~24–40GB | A100 40 GB+ (24 GB tight at BS 1) |”
https://github.com/huggingface/lerobot/blob/v0.6.1/docs/source/hardware_guide.mdx