DREAM Architecture: a Developmental Approach to Open-Ended Learning in Robotics

Abstract

Robots are still limited to controlled conditions, that the robot designer knows with enough details to endow the robot with the appropriate models or behaviors. Learning algorithms add some flexibility with the ability to discover the appropriate behavior given either some demonstrations or a reward to guide its exploration with a reinforcement learning algorithm. Reinforcement learning algorithms rely on the definition of state and action spaces that define reachable behaviors. Their adaptation capability critically depends on the representations of these spaces: small and discrete spaces result in fast learning while large and continuous spaces are challenging and either require a long training period or prevent the robot from converging to an appropriate behavior. Beside the operational cycle of policy execution and the learning cycle, which works at a slower time scale to acquire new policies, we introduce the redescription cycle, a third cycle working at an even slower time scale to generate or adapt the required representations to the robot, its environment and the task. We introduce the challenges raised by this cycle and we present DREAM (Deferred Restructuring of Experience in Autonomous Machines), a developmental cognitive architecture to bootstrap this redescription process stage by stage, build new state representations with appropriate motivations, and transfer the acquired knowledge across domains or tasks or even across robots. We describe results obtained so far with this approach and end up with a discussion of the questions it raises in Neuroscience.

Reducing Computational Cost During Robot Navigation and Human–Robot Interaction with a Human-Inspired Reinforcement Learning Architecture

“…Indeed, as designers of the system, we chose a representation (discretization of the output of a SLAM algorithm) adapted to the problem at hand (a navigation problem). However the context of this proposal is to build on the representation redescription framework [Doncieux et al, 2018[Doncieux et al, , 2020 to ultimately design systems that autonomously determine the representations adapted to the task. The modularity of the present architecture also enables to extend it to the continuous case by replacing tabular value functions with neural network implementations.…”

Explainability in Deep Reinforcement Learning

“…Equally important is to evaluate and explain other aspects in reinforcement learning, e.g. formally explaining the role of curriculum learning [82], quality diversity or other human-learning inspired aspects of open-ended learning [28,77,83]. Thus, more theoretic bases to serve explainable by design DRL are required.…”

“…This enables to capture the variations in the environment influenced by the agent's actions and thus, extrapolate explanations. SRL can be especially useful in RL for robotics and control [85,84,106,28,29], and can help to understand how the agent interprets the observations and what is relevant to learn to act, i.e., actionable or controllable features [62]. Indeed, the dimensionality reduction induced by SRL, coupled with the link to the control and possible disentanglement of variation factors, could be highly beneficial to improve our understanding capacity of the decisions made by RL algorithms using a state representation method [63].…”

Towards Continual Reinforcement Learning: A Review and Perspectives

“…Agents in ambitious open-ended learning settings, where both the agent and the designer do not know the task or the domain ahead of time, could also experience active sources of non-stationarity. For example, these settings may experience a changing state and action space over the course of an agent's lifetime while continuously generating creative behaviors (Doncieux et al, 2020). In active non-stationary environments, we can now assume that environment dynamics may vary in a Markovian way following the function p(z |s, a, z) as in (Ong et al, 2010).…”