Research map / Vision-language and multimodal / Robot action and navigation
Octopus: Embodied Vision-Language Programmer from Environmental Feedback
Jingkang Yang, Yuhao Dong, Shuai Liu, Bo Li, Ziyue Wang, ChenCheng Jiang, Haoran Tan, Jiamu Kang, Yuanhan Zhang, Kaiyang Zhou, Ziwei Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add RLEF (PPO with environmental-feedback reward model) after SFT (ablation) | Octopus · OctoGibson all tasks | task completion rate: 0.15 -> 0.18 | Improved |
| add RLEF after SFT (ablation) | Octopus · OctoGibson reasoning tasks | task completion rate: 0.13 -> 0.20 | Improved |
| add RLEF after SFT (ablation) | Octopus · OctoMC unseen tasks | task completion rate: 0.20 -> 0.20 | No clear effect |
| add RLEF after SFT (ablation) | Octopus · OctoGTA all | task completion rate: 0.18 -> 0.20 | Improved |
| randomize order of FPV and BEV image inputs | Octopus · OctoGibson | task performance | Hurt |
| reduce model size from 7B to 3B | Octopus · OctoGibson | task performance | Hurt |
| finetune language decoder in addition to connector (ablation) | Octopus · OctoGibson (60 tasks) | successful tasks: 4 -> 5 out of 60 | Improved |
| use code-specialized CodeLLaMA instead of general LLaMA as blind LLM | blind LLM planners · OctoGibson | executable code rate: 24% -> 92% | Improved |
| replace MPT backbone with CodeLLaMA in Octopus | Octopus · OctoGibson | output quality | Hurt |
| compare Octopus (SFT+RLEF) to GPT-4 and GPT-4V | Octopus · OctoGibson all | task completion rate: 0.18 vs 0.43 (GPT-4) | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- EmbodiedGPT — EmbodiedGPT: Vision-Language Pre-Training via Embodied Chain of Thought (NeurIPS 2023) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CaP-X: A Framework for Benchmarking and Improving Coding Agents for Robot Manipulation (ICML 2026 · similarity 0.90)
- Steve-Eye: Equipping LLM-based Embodied Agents with Visual Perception in Open Worlds (ICLR 2024 · similarity 0.89)
- SimWorld Studio: Automatic Environment Generation with Evolving Coding Agent for Embodied Agent Learning (NeurIPS 2026 · similarity 0.89)
- EmbodiedBench: Comprehensive Benchmarking Multi-modal Large Language Models for Vision-Driven Embodied Agents (ICML 2025 · similarity 0.88)
- Towards Reliable Code-as-Policies: A Neuro-Symbolic Framework for Embodied Task Planning (NeurIPS 2025 · similarity 0.88)
- VirtualEnv: A Platform for Embodied AI Research (AAAI 2026 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
