File size: 2,372 Bytes
c5a836f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
DW05-Robotwin
===============

Copyright 2026 Dexmal.

This release contains DW05 model weights and optional runtime components for
RobotWin-style world-action inference.  The package is distributed under the
Apache License 2.0 unless a component is explicitly attributed to an upstream
project below.

DW05 / Dexbotic
---------------

Project: DW05 world-action model for the Dexbotic runtime.
Release package: https://huggingface.co/Dexmal/DW05-Robotwin

The `model.pt` checkpoint is a DW05 trained checkpoint.  It includes DW05 model
parameters for the world-action runtime, including the trained video/action/MoT
weights and proprio encoder used by the released RobotWin policy.

Wan2.2 Components
-----------------

DW05 is built on and used with Wan2.2 video-generation components.

Upstream project: https://github.com/Wan-Video/Wan2.2
Upstream model: https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B
Upstream license: Apache License 2.0

When this bundle includes files under `vae/` or `text_encoder/`, those files are
runtime components compatible with the DW05 model and remain subject to the
upstream Apache License 2.0 terms and attribution requirements.

uMT5 Tokenizer / Text Components
--------------------------------

DW05 uses uMT5-compatible tokenization/text components for prompt encoding.

Upstream model: https://huggingface.co/google/umt5-xxl
Upstream license: Apache License 2.0

When this bundle includes files under `tokenizer/`, those tokenizer files remain
subject to the upstream Apache License 2.0 terms and attribution requirements.

RoboTwin / RobotWin Data And Evaluation
---------------------------------------

DW05-Robotwin is intended for RobotWin/RoboTwin-style data and evaluation.

Upstream project: https://github.com/RoboTwin-Platform/RoboTwin
Public dataset metadata checked at: https://huggingface.co/datasets/TianxingChen/RoboTwin2.0
Upstream license observed for the RoboTwin code and public dataset metadata: MIT License

Users should verify that any local datasets, task assets, or generated data used
with this checkpoint are licensed for their intended use.

No Warranty
-----------

This model package is provided on an AS IS basis, without warranties or
conditions of any kind.  Users are responsible for evaluating safety,
robustness, and legal suitability before redistribution, real-robot deployment,
or commercial use.