A recurrent spiking neural network is proposed that implements planning as probabilistic inference for finite and infinite horizon tasks. The architecture splits this problem into two parts: The stochastic transient firing of the network embodies the dynamics of the planning task. With appropriate injected input this dynamics is shaped to generate high-reward state trajectories. A general class of reward-modulated plasticity rules for these afferent synapses is presented. The updates optimize the likelihood of getting a reward through a variant of an Expectation Maximization algorithm and learning is guaranteed to convergence to a local maximum. We find that the network dynamics are qualitatively similar to transient firing patterns during planning and foraging in the hippocampus of awake behaving rats. The model extends classical attractor models and provides a testable prediction on identifying modulating contextual information. In a real robot arm reaching and obstacle avoidance task the ability to represent multiple task solutions is investigated. The neural planning method with its local update rules provides the basis for future neuromorphic hardware implementations with promising potentials like large data processing abilities and early initiation of strategies to avoid dangerous situations in robot co-worker scenarios.

No abstract

Autonomous robots need to interact with unknown, unstructured and changing environments, constantly facing novel challenges. Therefore, continuous online adaptation for lifelong-learning and the need of sample-efficient mechanisms to adapt to changes in the environment, the constraints, the tasks, or the robot itself are crucial. In this work, we propose a novel framework for probabilistic online motion planning with online adaptation based on a bio-inspired stochastic recurrent neural network. By using learning signals which mimic the intrinsic motivation signal cognitive dissonance in addition with a mental replay strategy to intensify experiences, the stochastic recurrent network can learn from few physical interactions and adapts to novel environments in seconds. We evaluate our online planning and adaptation framework on an anthropomorphic KUKA LWR arm. The rapid online adaptation is shown by learning unknown workspace constraints sample-efficiently from few physical interactions while following given way points.

To learn control policies in unknown environments, learning agents need to explore by trying actions deemed suboptimal. In prior work, such exploration is performed by either perturbing the actions at each time-step independently, or by perturbing policy parameters over an entire episode. Since both of these strategies have certain advantages, a more balanced trade-off could be beneficial. We introduce a unifying view on step-based and episode-based exploration that allows for such balanced trade-offs. This trade-off strategy can be used with various reinforcement learning algorithms. In this paper, we study this generalized exploration strategy in a policy gradient method and in relative entropy policy search. We evaluate the exploration strategy on four dynamical systems and compare the results to the established step-based and episode-based exploration strategies. Our results show that a more balanced trade-off can yield faster learning and better final policies, and illustrate some of the effects that cause these performance differences.

Despite recent progress in robot learning, it still remains a challenge to program a robot to deal with open-ended object manipulation tasks. One approach that was recently used to autonomously generate a repertoire of diverse skills is a novelty based Quality-Diversity (QD) algorithm. However, as most evolutionary algorithms, QD suffers from sampleinefficiency and, thus, it is challenging to apply it in real-world scenarios. This paper tackles this problem by integrating a neural network that predicts the behavior of the perturbed parameters into a novelty based QD algorithm. In the proposed Model-based Quality-Diversity search (M-QD), the network is trained concurrently to the repertoire and is used to avoid executing unpromising actions in the novelty search process. Furthermore, it is used to adapt the skills of the final repertoire in order to generalize the skills to different scenarios. Our experiments show that enhancing a QD algorithm with such a forward model improves the sample-efficiency and performance of the evolutionary process and the skill adaptation.