TP4: Adjustable Autonomy and Physical Embodied Intelligence


  • Fabio Patrizi (La Sapienza Università di Roma)
  • Barbara Caputo (Politecnico di Torino)

Spoke 2: Integrative AI
Spoke 3: Resilient AI
Spoke 4: Adaptive AI
Spoke 5: High quality AI
Spoke 7: Edge-exascale AI
Spoke 9: Green-aware AI
Spoke 10: Bio-socio-cognitive AI

Regulating autonomy of AI systems is a crucial challenge of future AI. TP4 aims at developing the science, techniques, and tools for adjustable autonomy for AI, possibly physical, systems, to provide them with the ability to autonomously operate, learn and understand when decisions are unsafe, questionable or unethical, and require human supervision.

The main research question TP4 aims to address is: “How to create autonomous AI systems able to operate and learn autonomously, and understand the limits of their autonomy, asking for human supervision, when appropriate?”

Adjustable autonomy is a form of intelligence providing agents with as much autonomy as possible, while retaining the ability to understand when they should autonomously ask for human help. Physical embodiment is the acquisition of full control of a physical device, including sensing and acting capabilities, and physical embodied intelligence refers to the challenges an agent must address when executing in a physical device.

The main TP4’s objectives are to:

  • Develop autonomous, self-adaptive and self-aware agents able to resort to external help;
  • Develop systems able to interact with the environment and other agents in a humane and predictive way;
  • Increase AI systems’ trustworthiness;
  • Inject embodiment-independent intelligence in agents;
  • Enable continual and autonomous learning of perceptual and semantic knowledge about the environment, by interacting with humans, machines, and other resources.