Meta tags:
description=;
keywords=Memory, Robotic Manipulation, Benchmark, Policy Design, Simulation, Real-world Experiments;

Headings (most frequently used words):

benchmark, policy, rmbench, memory, dependent, robotic, manipulation, with, insights, into, design, overview, mem, and, study, error, analysis, bibtex,

Text of the page (most frequently used words):
and (22), #memory (13), #manipulation (8), policy (7), the (7), university (6), benchmark (5), design (5), wang (5), rmbench (4), dependent (4), robotic (4), with (4), insights (4), into (4), chen (4), mem (4), module (4), have (3), hao (3), arxiv (3), planning (3), execution (3), subtask (3), task (3), policies (3), author (2), tianxing (2), yuran (2), mingleyang (2), qin (2), yan (2), shi (2), zixuan (2), yifan (2), zhang (2), yingsheng (2), kaixuan (2), yue (2), level (2), observations (2), key (2), frame (2), completion (2), existing (2), capabilities (2), tasks (2), that (2), over (2), real (2), world (2), systematic (2), evaluation (2), architectural (2), choices (2), performance (2), simulation (2), hong (2), kong (2), you, any, questions, please, contact, chentianxing2002, gmail, com, article, chen2026rmbench, title, others, journal, preprint, 2603, 01229, year, 2026, bibtex, error, analysis, study, comprises, linked, end, classifier, generates, high, subtasks, from, instructions, while, produces, low, actions, using, current, observation, fused, anchor, sliding, memories, diffusion, based, upon, stored, enable, iterative, until, pipeline, made, rapid, progress, recent, years, yet, most, approaches, give, limited, consideration, consequently, they, struggle, solve, require, reasoning, historical, maintaining, relevant, information, time, which, are, common, requirements, scenarios, although, several, aware, been, proposed, remains, underexplored, relationship, between, still, not, well, understood, address, this, gap, introduce, comprising, span, multiple, levels, complexity, enabling, further, propose, modular, explicit, components, designed, support, controlled, ablation, studies, through, extensive, experiments, identify, related, limitations, provide, empirical, how, influence, abstract, overview, code, equal, contribution, corresponding, mmlab, peking, psibot, science, technology, guangzhou, tsinghua, shenzhen, shanghai, jiao, tong, ping, luo, dong, yaodong, yang, yao, ruihai, renjing, hongcheng, under, review,

Text of the page (random words):
rmbench memory dependent robotic manipulation benchmark with insights into policy design rmbench memory dependent robotic manipulation benchmark with insights into policy design under review tianxing chen 1 yuran wang 2 3 mingleyang li 2 yan qin 4 hao shi 5 zixuan li 6 yifan hu 2 yingsheng zhang 5 kaixuan wang 1 yue chen 2 hongcheng wang 2 renjing xu 4 ruihai wu 2 yao mu 7 yaodong yang 2 3 hao dong 2 ping luo 1 1 mmlab the university of hong kong 2 peking university 3 psibot 4 the hong kong university of science and technology guangzhou 5 tsinghua university 6 shenzhen university 7 shanghai jiao tong university equal contribution corresponding author arxiv code overview 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 mem 0 policy mem 0 pipeline mem 0 comprises a planning module and an execution module linked by a subtask end classifier the planning module generates high level subtasks from task instructions observations and key frame memory while the execution module produces low level actions using the current observation the subtask and fused anchor and sliding memories in a diffusion based policy upon subtask completion a key frame is stored to enable iterative planning and execution until task completion benchmark and study error analysis bibtex article chen2026rmbench title rmbench memory dependent robotic manipulation benchmark with insights into policy design author chen tianxing and wang yuran and li mingleyang and qin yan and shi hao and li zixuan and hu yifan and zhang yingsheng and wang kaixuan and chen yue and others journal arxiv preprint arxiv 2603 01229 year 2026 if you have any questions please contact us at chentianxing2002 gmail com