RMBench: Memory-Dependent Robotic Manipulation Benchmark with Insights into Policy Design

2026-03-01

AI Summary

This work addresses the prevalent lack of systematic memory modeling in current robotic manipulation strategies, which struggle with real-world tasks requiring reliance on historical observations and long-term information retention. To bridge this gap, we introduce RMBench, the first standardized simulation benchmark tailored for memory-dependent manipulation, featuring nine tasks spanning varying levels of memory complexity. We further propose Mem-0, a modular policy architecture that explicitly integrates dedicated memory components. Through multi-level task design, ablation studies, and real-robot validation, our analysis systematically uncovers the critical relationship between policy architecture and memory performance. The findings not only expose fundamental limitations of existing approaches in handling memory-intensive tasks but also provide reproducible architectural guidelines for enhancing memory capabilities in robotic manipulation systems.

Technology Category

• Intelligent Robots: Manipulation
• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Robotic manipulation policies have made rapid progress in recent years, yet most existing approaches give limited consideration to memory capabilities. Consequently, they struggle to solve tasks that require reasoning over historical observations and maintaining task-relevant information over time, which are common requirements in real-world manipulation scenarios. Although several memory-aware policies have been proposed, systematic evaluation of memory-dependent manipulation remains underexplored, and the relationship between architectural design choices and memory performance is still not well understood. To address this gap, we introduce RMBench, a simulation benchmark comprising 9 manipulation tasks that span multiple levels of memory complexity, enabling systematic evaluation of policy memory capabilities. We further propose Mem-0, a modular manipulation policy with explicit memory components designed to support controlled ablation studies. Through extensive simulation and real-world experiments, we identify memory-related limitations in existing policies and provide empirical insights into how architectural design choices influence memory performance. The website is available at https://rmbench.github.io/.

Problem

• robotic manipulation
• memory-dependent tasks
• policy evaluation
• architectural design
• benchmark

Innovation

• memory-dependent manipulation
• robotic benchmark
• modular policy
• memory architecture
• ablation study