AI TrackerFrontier desk · AEST
Back to news
normalaitech

Multimodal Embodied Agents: A Unified Capability-centric Perspective from Computer-Use to Robot-Use

arXiv cs.AI · 22 Oct 2026, 03:52 AEST

Multimodal embodied agents (MMEAs) are defined as goal-directed systems that integrate multimodal task reasoning with physical actions. The PAPAV framework introduces five key functions to evaluate and compare these agents with multimodal agents and robotic systems.

Key facts

Read aloud

Multimodal embodied agents MMEAs are defined as goal-directed systems that integrate multimodal task reasoning with physical actions. The PAPAV framework introduces five key functions to evaluate and compare these agents with multimodal agents and robotic systems. Multimodal agents MMAs sustain interaction through reasoning, memory, tools, and feedback, most visibly as computer-use agents, while robotic systems RSs couple sensing and actuation under physical dynamics. We define multimodal embodied agents MMEAs as goal-directed systems that couple multimodal task reasoning with physical action and revise decisions from the resulting feedback. Five physical constraints, from partial observability to unverifiable outcomes, leave fewer choices fixed in advance and less room to reverse errors. Across sixty four benchmarks, explicit evaluation centers on Act; only five assess Anticipate and three assess Verify, leaving both largely hidden behind task success. PAPAV therefore provides a common basis for designing reliable physical agents and measuring progress across all five capabilities rather than task success alone.

Sources in this cluster