PEPA: a Persistently Autonomous Embodied Agent with Personalities
A quadruped robot autonomously navigated a multi-floor office building for days, guided by five distinct personality prototypes.
A research team led by Kaige Liu has introduced PEPA (Persistently Autonomous Embodied Agent with Personalities), a novel framework that uses personality as an intrinsic organizational principle to enable long-term robotic autonomy. Published on arXiv, the work addresses a core limitation in robotics: current embodied agents remain dependent on externally scripted objectives, making them impractical for sustained deployment in unstructured, dynamic settings. PEPA proposes that personality traits, analogous to genotypic biases in biology, can provide the internal drive for an agent to autonomously generate and refine its own goals, leading to self-sustaining behavioral evolution without constant human supervision.
The team realized this vision with a three-layer cognitive architecture. The high-level Sys3 synthesizes personality-aligned goals and refines them through episodic memory and daily self-reflection. Sys2 performs deliberative reasoning to translate these goals into executable plans, while Sys1 handles sensorimotor interaction and experience recording. The system was validated through real-world deployment of a quadruped robot in a multi-floor office building. Operating for extended periods without fixed task specifications, the robot successfully navigated elevators and explored environments, autonomously arbitrating between external user requests and its internal, personality-driven motivations. Quantitative analysis across five distinct personality prototypes confirmed stable, trait-aligned behaviors, marking a significant step toward creating robots capable of persistent, unsupervised operation.
- Uses a three-layer cognitive architecture (Sys1, Sys2, Sys3) where personality in Sys3 autonomously generates and refines goals.
- Successfully deployed on a real quadruped robot that navigated a multi-floor office building and used elevators without predefined tasks.
- Demonstrated five distinct, stable personality prototypes that guide the agent's long-term behavior and decision-making.
Why It Matters
Enables robots to operate for days in dynamic real-world environments without human intervention, moving beyond scripted tasks.