Katsunari Shibata

About

Katsunari Shibata is a pioneering robotics and artificial intelligence researcher whose work has fundamentally advanced the field of end-to-end reinforcement learning (RL) for autonomous robotic systems. Over more than two decades, Shibata has championed a deceptively simple yet powerful framework: coupling raw sensory inputs — particularly unprocessed visual signals from CCD cameras — directly with neural networks trained via reinforcement learning, allowing robots to autonomously acquire complex behaviors without human-engineered feature extraction. His most celebrated contributions include demonstrating that real mobile robots could learn visually guided tasks such as box-pushing purely from raw pixel inputs, and showing that higher cognitive functions — including prediction, contextual reasoning, and flexible decision-making — can *emerge* organically through end-to-end learning with recurrent neural networks. Long before Google DeepMind's headline-grabbing results, Shibata's group was already exploring this foundational paradigm, a fact he has documented across influential publications accumulating dozens of citations. Shibata's broader vision, articulated in works on artificial general intelligence, argues that massively parallel, cohesively integrated neural learning systems represent the most promising path toward human-like robotic intelligence — a perspective that has proven remarkably prescient given contemporary deep learning trends.

Research Focus

Key Achievements

Top Papers

1. 1
   Acquisition of box pushing by direct-vision-based reinforcement learning

   23 citations · 2003
2. 2
   Contextual Behaviors and Internal Representations Acquired by Reinforcement Learning with a Recurrent Neural Network in a Continuous State and Action Space Task

   20 citations · 2009
3. 3
   Emergence of Intelligence through Reinforcement Learning with a Neural Network

   14 citations · 2011
4. 4
   Functions that Emerge through End-to-End Reinforcement Learning - The Direction for Artificial General Intelligence -

   14 citations · 2017
5. 5
   Emergence of Prediction by Reinforcement Learning Using a Recurrent Neural Network

   10 citations · 2010
6. 6
   Emergence of flexible prediction-based discrete decision making and continuous motion generation through actor-Q-learning

   10 citations · 2013
7. 7
   Acquisition of Flexible Image Recognition by Coupling of Reinforcement Learning and a Neural Network

   9 citations · 2009
8. 8
   Robotic batting system — an architecture for learning and dynamic pattern generation

   8 citations · 2000
9. 9
   Learning of Action Generation from Raw Camera Images in a Real-World-Like Environment by Simple Coupling of Reinforcement Learning and a Neural Network

   7 citations · 2009
10. 10
    Application of direct-vision-based reinforcement learning to a real mobile robot

    7 citations · 2002

Key Collaborators

Contact & Links