AndroTMem: From Interaction Trajectories to Anchored Memory in Long-Horizon GUI Agents

Dimensions overall score 8.0

• AndroTMem-Bench comprises 1,069 tasks with 34,473 interaction steps, averaging 32.1 steps per task with a maximum of 65 steps.technical 100%
• As interaction sequences grow longer in GUI agents, performance drops are driven mainly by within-task memory failures rather than isolated perception errors or local action mistakes.result 90%
• Anchored State Memory (ASM) consistently outperforms full-sequence replay and summary-based baselines across multiple settings and 12 evaluated GUI agents.result 90%
• Anchored State Memory improves Task Complete Rate by 5% to 30.16% compared to baseline methods.result 90%
• AndroTMem-Bench is designed to enforce strong step-to-step causal dependencies, making sparse yet essential intermediate states decisive for downstream actions.technical 85%
• Anchored State Memory represents interaction sequences as a compact set of causally linked intermediate-state anchors to enable subgoal-targeted retrieval and attribution-aware decision making.method 85%
• Replaying full interaction sequences in GUI agents is redundant and amplifies noise, while summaries often erase dependency-critical information and traceability.limitation 80%
• Effective interaction memory under prevailing paradigms for long-horizon GUI agents remains under-explored.limitation 80%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card