Yuquan-Wang/d9_rough-walking-2025-06-07_05-23-21-episode-20000

IsaacLab Training Output

Model Checkpoints

• Final model: model_9999.pt

Checkpoint 2000: model_2000.pt- Checkpoint 5000: model_5000.pt- Checkpoint 8000: model_8000.pt- Checkpoint 9000: model_9000.pt- Checkpoint 10000: model_10000.pt- Checkpoint 11000: model_11000.pt- Checkpoint 12000: model_12000.pt- Checkpoint 13000: model_13000.pt- Checkpoint 14000: model_14000.pt- Checkpoint 15000: model_15000.pt- Checkpoint 16000: model_16000.pt- Checkpoint 17000: model_17000.pt- Checkpoint 18000: model_18000.pt- Checkpoint 19000: model_19000.pt- Checkpoint 19999: model_19999.pt

Exported Models

Optimized models for deployment are available in the exported/ directory:

• policy.pt: Optimized PyTorch model
• policy.onnx: ONNX format model for cross-platform deployment

Training Configuration

See the params/ directory for training configuration files.

Training Logs

Training logs are available in the logs/ directory.

Demo Videos

Demo videos are available in the videos/ directory.

Usage

Using PyTorch Checkpoints

import torch
model = torch.load('path/to/model.pt')

Using Exported Models

# PyTorch
import torch
model = torch.load('exported/policy.pt')

# ONNX (requires onnxruntime)
import onnxruntime as ort
session = ort.InferenceSession('exported/policy.onnx')

Notes

This version limits the lateral velocity to +- 0.1 m/s.