RAVEN: Long-Horizon Reasoning and Navigation with a Visuo-Spatial-Temporal Memory

[Paper] [Website] [Dataset] [BibTeX]

Yixun Hu*, Zhicheng Zheng*, Lihan Zha, Chunwei Xing, Rajdeep Singh, Omar Hossain, Antonio Loquercio, and Dhruv Shah (* Equal contribution)

Note (2026-02-04): If you find any serious bugs or issues, please open an issue and we will address them in a timely manner. Contributions and code improvements are welcome via pull requests. Thank you!

Setup

bash docs/raven_setup.sh raven
conda activate raven

If use open-source VLM models

curl -fsSL https://ollama.com/download/ollama-{Your-Version}.tgz  |  tar zx -C ~/.cache/ollama/

# for example
curl -fsSL https://ollama.com/download/ollama-linux-amd64.tgz  |  tar zx -C ~/.cache/ollama/

Running Benchmarks

For step-by-step instructions on running RAVEN, please refer to the following docs:

1. RAVEN-QA: A challenging benchmark for robot long-term memory and navigation with visual information recall, visual reasoning, small object retrieval, and so forth.

2. NaVQA A question-answering dataset based on CODa dataset.

3. FindingDory A Long-term loco-manipulation dataset and object retrieval benchmark based on Habitat.

Acknowledgements

• This work is inspired by previous papers or implementations: ReMEmbR, VLFM, and QQMM-embed. We greatly thank these pioneering works. In particular, this project was originally inspired by and partially built upon the codebase of: https://github.com/NVIDIA-AI-IOT/remembr (ReMEmbR: Building and Reasoning Over Long-Horizon Spatio-Temporal Memory for Robots, Abrar Anwar, John Welsh, Joydeep Biswas, Soha Pouya, Yan Chang. [ICRA 2025])

• Datasets: 1. Yadav, K. et al. FindingDory: A Benchmark to Evaluate Memory in Embodied Agents. arXiv:2506.15635, 2025. https://arxiv.org/abs/2506.15635 2. Zhang, A. et al. UT Campus Object Dataset (CODa). Texas Data Repository, Draft Version, 2023. DOI: 10.18738/T8/BBOQMV.

BibTeX

@article{hu2026raven,
  title={RAVEN: Long-Horizon Reasoning \& Navigation with a Visuo-Spatio-Temporal Memory},
  author={Hu, Yixun and Zheng, Zhicheng and Zha, Lihan and Xing, Chunwei and Singh, Rajdeep and Hossain, Omar and Loquercio, Antonio and Shah, Dhruv},
  journal={arXiv preprint arXiv:2606.25206},
  year={2026}
}