Thrilled to announce that our paper E3VS-BENCH has been accepted to #ECCV2026!
I'm deeply grateful to all the co-authors and collaborators who contributed to this project.
See you in Malmö!
#ECCV#ECCV2026
By unifying prediction and control, NavWAM acts as a closed-loop policy out of the box. In evaluations, it outperforms planning-based world models without needing test-time action search, matches a much larger 7B VLA policy, and transfers successfully to real mobile robots.
Excited to share E3VS-Bench, a benchmark for Embodied 3D Visual Search!
Many existing visual search benchmarks can be solved from a limited set of viewpoints or through planar navigation alone.
However, real-world visual search often requires richer viewpoint control: looking up or down, moving higher or lower, and inspecting objects from specific viewing angles to reveal information hidden by occlusions or object geometry.
To study this challenge, we introduce E3VS-Bench, a benchmark for viewpoint-dependent active perception in photorealistic 3D environments.
🔹 2,014 visual search episodes
🔹 Built on 99 real-world 3DGS scenes
🔹 Full 5-DoF viewpoint control
🔹 Human and frontier VLM evaluation
Our results reveal a substantial gap between humans and current models, highlighting the difficulty of active perception in 3D environments.
🌍 Project Page: https://t.co/9opDeSpvdV
📄 Paper: https://t.co/S6RbcuObs7
💻 Code & Dataset: https://t.co/JHdHQlgKWv
#CoRL2025 に、M1勝又の論文が採択されました。修士以下のCoRL採択は2年連続です。
採択率は35.77%でした。
K. Katsumata, Y. Iioka, N. Hosomi, T. Misu, K. Yamada, K. Sugiura, GENNAV: Polygon Mask Generation for Generalized Referring Navigable Regions
本田技研との共同研究成果です。
[1/6] Recent models like DUSt3R generalize well across viewpoints, but performance drops on aerial-ground pairs.
At #CVPR2025, we propose AerialMegaDepth (https://t.co/tDGMVXAFa7), a hybrid dataset combining mesh renderings with real ground images (MegaDepth) to bridge this gap.
Our paper “LegalViz: Legal Text Visualization by Text To Diagram Generation” is accepted at #NAACL2025 main!!
法律と自然言語処理に関する主著論文がNAACLにアクセプトされました。言語処理学会にて同内容の発表を行いますので、ぜひ来てください!
Introducing MASt3R-SLAM, the first real-time monocular dense SLAM with MASt3R as a foundation.
Easy to use like DUSt3R/MASt3R, from an uncalibrated RGB video it recovers accurate, globally consistent poses & a dense map.
With @eric_dexheimer*, @AjdDavison (*Equal Contribution)
We will make 2 presentations in #IROS2024
1. Answerability Fields: Answerable Location Estimation via Diffusion Models
From bird-view to heat-map via diffusion process, our model predicts "Where the agent should go next?" to get the best viewpoint to answer embodied questions!
2. Map-based Modular Approach for Zero-shot Embodied Question Answering
We present a map-based modular approach to EQA, enabling real-world robots to explore and map unknown environments with a help of the foundation models. We deploy an automobile robot for EQA!
#IROS2024