Agentic Robotics: Redefining Robot Perception, Planning, Control, and Multi-Agent Autonomy through Agentic Super Intelligence
Agentic Artificial Intelligence (Agentic AI) is emerging as a system-level paradigm that reorganizes robotic autonomy beyond conventional perception-planning-control pipelines. This report traces the evolution from autonomy to agency and introduces a layered capability model connecting machine learning, generative AI, embodied foundation models, AI agents, and coordinated multi-agent systems. We define "Agentic AI in robotics" at the system level and synthesize an architecture integrating persistent perceive-reason-act-verify reflect loops, structured memory, tool invocation, and safety gates. Furthermore, this study examines agentic perception through goal-conditioned sensing and Vision-Language-Action alignment; agentic planning through goal decomposition, multimodal world modeling, and reflective replanning; and agentic control through semantic-to-physical translation, adaptive execution, verification, recovery, and multi-robot coordination. Existing robotics benchmarks are reframed using metrics for task competence, persistence, recovery, generalization, continual learning, coordination, latency, and deployability. Key limitations include hallucinated reasoning, memory drift, sim-to-real gaps, computational overhead, integration debt, and governance risks. Future agentic robotics will increasingly combine neuro-symbolic reasoning, physics-aware models, calibrated autonomy, formal verification, and governed multi-agent collaboration, enabling scalable, explainable, resilient, and trustworthy robotic systems.
Authors
- Manoj Karkee (ORCID: https://orcid.org/0000-0001-5337-4848)
- Ranjan Sapkota (ORCID: https://orcid.org/0000-0002-5417-6744)
- Seyedali Mirjalili (ORCID: https://orcid.org/0000-0002-1443-9458)
Publication Details
- Journal
- Figshare
- Published
- 2026-11-21
- DOI
- https://doi.org/10.6084/m9.figshare.35083633
- Primary Topic
- Robotics and Automated Systems
- Type
- preprint