Summary Intelligent animals devote much time and energy to exploring and obtaining information, but the underlying mechanisms are poorly understood. We review recent developments on this topic that have emerged from the traditionally separate fields of machine learning, eye movements in natural behavior, and studies of curiosity in psychology and neuroscience. These studies show that exploration may be guided by a family of mechanisms that range from automatic biases toward novelty or surprise, to systematic search for learning progress and information gain in curiosity-driven behavior. In addition, eye movements reflect visual information search in multiple conditions and are amenable for cellular-level investigations, suggesting that the oculomotor system is an excellent model system for understanding information sampling mechanisms.

a b s t r a c tWe introduce the Self-Adaptive Goal Generation Robust Intelligent Adaptive Curiosity (SAGG-RIAC) architecture as an intrinsically motivated goal exploration mechanism which allows active learning of inverse models in high-dimensional redundant robots. This allows a robot to efficiently and actively learn distributions of parameterized motor skills/policies that solve a corresponding distribution of parameterized tasks/goals. The architecture makes the robot sample actively novel parameterized tasks in the task space, based on a measure of competence progress, each of which triggers low-level goal-directed learning of the motor policy parameters that allow to solve it. For both learning and generalization, the system leverages regression techniques which allow to infer the motor policy parameters corresponding to a given novel parameterized task, and based on the previously learnt correspondences between policy and task parameters.We present experiments with high-dimensional continuous sensorimotor spaces in three different robotic setups: (1) learning the inverse kinematics in a highly-redundant robotic arm, (2) learning omnidirectional locomotion with motor primitives in a quadruped robot, and (3) an arm learning to control a fishing rod with a flexible wire. We show that (1) exploration in the task space can be a lot faster than exploration in the actuator space for learning inverse models in redundant robots; (2) selecting goals maximizing competence progress creates developmental trajectories driving the robot to progressively focus on tasks of increasing complexity and is statistically significantly more efficient than selecting tasks randomly, as well as more efficient than different standard active motor babbling methods; (3) this architecture allows the robot to actively discover which parts of its task space it can learn to reach and which part it cannot.

No abstract

Abstract-Intelligent adaptive curiosity (IAC) was initially introduced as a developmental mechanism allowing a robot to self-organize developmental trajectories of increasing complexity without preprogramming the particular developmental stages. In this paper, we argue that IAC and other intrinsically motivated learning heuristics could be viewed as active learning algorithms that are particularly suited for learning forward models in unprepared sensorimotor spaces with large unlearnable subspaces. Then, we introduce a novel formulation of IAC, called robust intelligent adaptive curiosity (R-IAC), and show that its performances as an intrinsically motivated active learning algorithm are far superior to IAC in a complex sensorimotor space where only a small subspace is neither unlearnable nor trivial. We also show results in which the learnt forward model is reused in a control scheme. Finally, an open source accompanying software containing these algorithms as well as tools to reproduce all the experiments presented in this paper is made publicly available.