AgenticRoboRL-Qwen3.5-9B

Two checkpoints of Qwen/Qwen3.5-9B post-trained as a robot planner.

📝 Paper: AgenticRoboRL: Bringing Coding-Agent Post-Training to Agentic Robotics · 💻 Code: L-CodingSpace/AgenticRoboRL

Both folders are complete Hugging Face model directories (bf16 weights, tokenizer and processor files).

from transformers import AutoProcessor, Qwen3_5ForConditionalGeneration

repo = "ciciciciliu/AgenticRoboRL-Qwen3.5-9B"
model = Qwen3_5ForConditionalGeneration.from_pretrained(repo, subfolder="rl-v006", torch_dtype="auto")
processor = AutoProcessor.from_pretrained(repo, subfolder="rl-v006")

Citation

@article{liu2026agenticroborl,
  title   = {AgenticRoboRL: Bringing Coding-Agent Post-Training to Agentic Robotics},
  author  = {Liu, Xinxin and Xu, Bingxin and Bao, Wenrui and Li, Ming and Bao, Wentao and Chen, Chen and Shang, Yuzhang},
  journal = {arXiv preprint arXiv:2610.28589},
  year    = {2026}
}