Benchmark Radar
AI BENCHMARK PROFILE

ReMemBench

Robotics & Autonomous SystemsRobotics & Embodied IntelligencePRISM team

ReMemBench is a benchmark with eight diverse household manipulation tasks across four categories of short-term memory, designed to evaluate memory mechanisms in visuomotor policies.

Released
2026-06-15
Readiness
Inspectable
Primary field
Robotics & Autonomous Systems

Why it matters

ReMemBench addresses the lack of systematic evaluation for short-term memory in visuomotor control, providing a standardized protocol to compare memory-augmented policies and guide development for long-horizon tasks.

Motivation

Many robotic tasks require short-term memory, whether it's retrieving an object that's no longer visible or turning off an appliance after a set period.

Primary resources

Benchmark Radar records only publicly supported details and links back to primary sources for verification.