Robotics
Asimov
ONNX
ishneet commited on
Commit
4df2eb3
·
verified ·
1 Parent(s): c842525

Add how to run and inputs/outputs to the model card

Browse files

Adds a "How to run" section with the one-command viewer (npm run hf Menlo/asimov1-locomotion-0818, no GPU needed) and an "Inputs and outputs" section describing the 78 observations and 23 joint actions.

Files changed (1) hide show
  1. README.md +30 -1
README.md CHANGED
@@ -10,4 +10,33 @@ license: bsd-3-clause
10
  - `policy.onnx` is the exported policy for inference, producing joint-position actions.
11
  - `agent.yaml` records the actor-critic and AMP training settings, including the motion data configuration.
12
  - `env.yaml` records the simulation and locomotion task settings, including observations, actions, commands, and rewards.
13
- - Training and evaluation code: [menloresearch/isaac_asimov](https://github.com/menloresearch/isaac_asimov).
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
10
  - `policy.onnx` is the exported policy for inference, producing joint-position actions.
11
  - `agent.yaml` records the actor-critic and AMP training settings, including the motion data configuration.
12
  - `env.yaml` records the simulation and locomotion task settings, including observations, actions, commands, and rewards.
13
+ - Training and evaluation code: [menloresearch/isaac_asimov](https://github.com/menloresearch/isaac_asimov).
14
+
15
+ ## How to run
16
+
17
+ Run the policy in a browser simulation of Asimov 1 with
18
+ [humanoid-policy-viewer](https://github.com/menloresearch/humanoid-policy-viewer).
19
+ It needs Node, git and bash, and no GPU:
20
+
21
+ ```bash
22
+ git clone https://github.com/menloresearch/humanoid-policy-viewer
23
+ cd humanoid-policy-viewer
24
+ npm run hf Menlo/asimov1-locomotion-0818
25
+ ```
26
+
27
+ This downloads the policy, starts the viewer and opens it in your browser.
28
+ Drive the robot with the velocity sliders, push it, or run the benchmark suite.
29
+
30
+ ## Inputs and outputs
31
+
32
+ - **Input:** 78 values at 50 Hz: base angular velocity, projected gravity,
33
+ velocity command (`vx, vy, wz`), joint positions and velocities, and the
34
+ previous action. All of these are available on the real robot.
35
+ - **Output:** 23 joint position actions (legs, waist and arms; the neck is not
36
+ driven). Each target is `default_joint_pos + action_scale * action`, with
37
+ both values in `env.yaml`.
38
+ - Trained in Isaac Lab (PhysX); the viewer simulates it in MuJoCo.
39
+
40
+ The full observation order and joint list are in the viewer's
41
+ [supported policies](https://github.com/menloresearch/humanoid-policy-viewer/blob/main/docs/huggingface.md#supported-policies)
42
+ doc.