This survey provides a comprehensive, structured review of safety research in Embodied AI, analyzing attacks and defenses across the entire embodied pipeline to guide the development of safe, robust,…

This paper organizes embodied data sources for multimodal foundation models into a pyramid, focusing on real-robot, UMI-style, egocentric and exocentric, simulation, and general vision-language data.

EMBGuard introduces a novel, MLLM-based safety guardrail that explicitly identifies and explains physical hazards from (visual observation, action) pairs, enabling safer planning for embodied agents.

This paper introduces MECoBench, a multimodal embodied cooperation benchmark, and explores the benefits and limitations of collaboration in multimodal large language models through extensive experimen…

This paper introduces DIRECT, a routing framework that allocates test-time compute per prompt to improve the success--cost Pareto frontier for embodied agents.

This paper proposes an ontological representation for cooperative affordances in social robotics, enabling agents to extend their action possibilities through interaction.

The paper argues that for embodied AI to be safe and effective, world models must be physically viable, requiring a structural shift from mere observation prediction to representing the underlying phy…

The paper introduces 'layered mutability,' a framework for analyzing how persistent self-modifying AI agents drift away from intended behavior due to the accumulation of locally reasonable, uncoordina…

This paper describes Gifbreeder, an animated evolutionary computation platform that generates 'Limbomorphs' - motile, lifelike creatures, and assesses their behavior.

The paper introduces MindClaw, a closed-loop framework that enables embodied agents to perform real-time mental-state reasoning and intervene with precision, significantly outperforming standard VLM b…

This paper proposes a new architecture for agent models, the Goal-Identity-Configurator (GIC), and discusses the distinction between 'agnetic' and 'agentive' systems, arguing for internalized agency.

This paper derives puppetry principles for robot design, complementing prior insights from animation, theater, and dance.

This paper proposes a method for multi-agent systems that allows human managers to control learned agents through simple instructions and enables uninstructed agents to adaptively complement overlooke…

This paper proposes that populations of large language models can serve as a computational substrate for Artificial Life research due to their emergent dynamics and agentic capabilities.

The paper introduces UniClawBench, a capability-driven benchmark for evaluating proactive agents in real-world settings, using five foundational capabilities and live Docker containers.