Adaptive robotics achieved tremendous progress during the last few years (see Nolfi (2021) for an introduction and review). The term adaptive robotics refers to methods which permit the design of robots capable of developing their skills autonomously through an evolutionary and/or learning process. It focuses on approaches requiring minimal human intervention in which the behavior displayed by the robots and the control rules producing such behavior are discovered by an adaptive process automatically on the basis of a reward or fitness function which rates how well the robot is doing. It focuses on end-to-end learning, i.e. on systems which receive as input directly the state of robot’s sensors and determine directly the state of the robot’s actuators, without involving any type of hand-designed pre-processing. Finally, it focuses on model-free methods, i.e. on systems which do not have an internal model of the environment, or in which the internal model is acquired automatically during the adaptation process. In this article I will review the major advances and the research challenges.
Progress and challenges in adaptive robotics
Nolfi S.
2022
Abstract
Adaptive robotics achieved tremendous progress during the last few years (see Nolfi (2021) for an introduction and review). The term adaptive robotics refers to methods which permit the design of robots capable of developing their skills autonomously through an evolutionary and/or learning process. It focuses on approaches requiring minimal human intervention in which the behavior displayed by the robots and the control rules producing such behavior are discovered by an adaptive process automatically on the basis of a reward or fitness function which rates how well the robot is doing. It focuses on end-to-end learning, i.e. on systems which receive as input directly the state of robot’s sensors and determine directly the state of the robot’s actuators, without involving any type of hand-designed pre-processing. Finally, it focuses on model-free methods, i.e. on systems which do not have an internal model of the environment, or in which the internal model is acquired automatically during the adaptation process. In this article I will review the major advances and the research challenges.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.