ActiveArena benchmarks robot vision and memory in manipulation tasks
ActiveArena: Benchmarking and Understanding Active Perception in Robotic Manipulation
RoboticsArtificial Intelligence
Summary
Robots need to explore and remember information to handle complex tasks well. The authors created ActiveArena, a simulator and benchmark with many tasks that require robots to actively seek information and use memory. Their benchmark tests how robots handle new scenes and challenges by measuring how well they explore and decide what to remember. They also studied different memory and planning strategies to find what helps robots perform better on unfamiliar tasks. This work helps researchers build robots that see and think actively instead of passively watching.
What this means in practice
- •For robotics engineers: Evaluate and improve robot systems that need active scene exploration and memory for manipulation tasks in varied environments.
- •For industrial automation teams: Test robotic handling solutions to ensure reliable object interaction when dealing with new or changing factory layouts.
Authors
Yibo Li, Enshen Zhou, Rui Chen, Yanjun Ding, Mengzhen Liu, Yi Han, Jiabo Zhan, Lipeng Wang, Shanghang Zhang, Lu Sheng
Abstract
Active perception and manipulation are crucial for robots to interact with complex scenes. Existing benchmarks struggle to evaluate how robots effectively acquire and maintain information in memory in an active manner. To this end, we introduce ActiveArena-Sim, an active-perception simulator with controllable viewpoints and large-scale workspaces as the foundation. Built on this, we propose ActiveArena-Bench, which comprises 35 tasks across 5 fine-grained categories, covering visual exploration and interactive information acquisition. Each task is difficult to solve from passive observations alone, requiring multi-round evidence acquisition and memory-based reasoning. The benchmark provides rich memory annotations, standardized training data, and ID/OOD protocols featuring disjoint scenes, unseen distractor configurations, and novel backgrounds. Moreover, we present ActiveArena-VLA, a modular suite of 13 vision-language-action configurations for controlled studies of memory writing, memory capacity, proprioceptive state, subtask supervision, and high-level planning in active perception. Benchmark results reveal a substantial ID-OOD gap: uniform memory sampling, increased memory capacity under reliable write policies, proprioceptive inputs, and subtask supervision improve OOD generalization, while planner-guided memory management and decision-making achieve performance close to the best-performing configuration using only sparse memory. ActiveArena thus provides a unified testbed to develop and diagnose models for active perception and manipulation.