[ICML2026] ACTIVE-O3: Empowering Multimodal Large Language Models with Active Perception via GRPO
-
Updated
Apr 30, 2026
[ICML2026] ACTIVE-O3: Empowering Multimodal Large Language Models with Active Perception via GRPO
Code for the paper: "Active Vision Might Be All You Need: Exploring Active Vision in Bimanual Robotic Manipulation"
Code for NeurIPS 2023 paper "Active Vision Reinforcement Learning with Limited Visual Observability"
GapFlyt: Active Vision Based Minimalist Structure-less Gap Detection For Quadrotor Flight
[RA-L'26] Online Visibility-Aware Replanning for Occlusion-Free Aerial Scanning
CanViT: Toward Active-Vision Foundation Models (NeurIPS 2026). Model, pretraining, task specialization and evaluation code.
Bebop 2 custom firmware
Code basis for the paper "Monitoring and Adapting the Physical State of a Camera for Autonomous Vehicles" (2023)
A one-shot method for selecting next best view for active object recognition
Active object localization framework incorporating the bio-plausible mechanisms of foveation and saccades for improved and resilient weakly supervised object localization
The code release of "Real-time Active Vision for a Humanoid Soccer Robot Using Deep Reinforcement Learning" paper, ICAART 2021
An active vision system on the PR2 humanoid robot to dynamically detect objects via the head and arm cameras
Head-tracking active vision platform for humanoid robots — PICO 4 → dual-axis gimbal → D415 camera. 3D-printable mounts & URDF/SDF simulation models included.
A dataset for testing next best view methods as a part of active vision systems.
An active vision system which builds a 3D environment map autonomously using visual attention mechanisms.
Official implementation and analysis of the Glimpse Prediction Networks.
电鸡哨兵|YOLO + 强化学习的电动自行车烟火感知与主动巡视研究。已实现场景微调、视频检测、连续帧告警评测与公开权重/标注;强化学习模块协作开发中,待合入。
Official project website for DAVIS, a depth-only active-vision framework for humanoid soccer skills (arXiv:2609.28175).
This repository stores the project development of a ViZDOOM visual learning environment, a project developed by Rhys Stewart as part of the individual 4th year project at Glasgow University
TRACE: computer-vision research for calibrated change evidence and learned re-observation, with experiments and an interactive local studio.
To associate your repository with the active-vision topic, visit your repo's landing page and select "manage topics."