Recently, AlphaZero has achieved landmark results in deep reinforcement learning, by providing a single self-play architecture that learned three different games at super human level. AlphaZero is a large and complicated system with many parameters, and success requires much compute power and fine-tuning. Reproducing results in other games is a challenge, and many researchers are looking for ways to improve results while reducing computational demands. AlphaZero's design is purely based on self-play and makes no use of labeled expert data or domain specific enhancements; it is designed to learn from scratch. We propose a novel approach to deal with this cold-start problem by employing simple search enhancements at the beginning phase of self-play training, namely Rollout, Rapid Action Value Estimate (RAVE) and dynamically weighted combinations of these with the neural network, and Rolling Horizon Evolutionary Algorithms (RHEA). Our experiments indicate that most of these enhancements improve the performance of their baseline player in three different (small) board games, with especially RAVE based variants playing strongly.

This paper proposes a Deep Reinforcement Learning (DRL) approach for solving knapsack problem. The proposed method consists of a state aggregation step based on tabular reinforcement learning to extract features and construct states. The state aggregation policy is applied to each problem instance of the knapsack problem, which is used with Advantage Actor Critic (A2C) algorithm to train a policy through which the items are sequentially selected at each time step. The method is a constructive solution approach and the process of selecting items is repeated until the final solution is obtained. The experiments show that our approach provides close to optimal solutions for all tested instances, outperforms the greedy algorithm, and is able to handle larger instances and more flexible than an existing DRL approach. In addition, the results demonstrate that the proposed model with the state aggregation strategy not only gives better solutions but also learns in less timesteps, than the one without state aggregation.

We study reinforcement learning in settings where sampling an action from the policy must be done concurrently with the time evolution of the controlled system, such as when a robot must decide on the next action while still performing the previous action. Much like a person or an animal, the robot must think and move at the same time, deciding on its next action before the previous one has completed. In order to develop an algorithmic framework for such concurrent control problems, we start with a continuous-time formulation of the Bellman equations, and then discretize them in a way that is aware of system delays. We instantiate this new class of approximate dynamic programming methods via a simple architectural extension to existing value-based deep reinforcement learning algorithms. We evaluate our methods on simulated benchmark tasks and a large-scale robotic grasping task where the robot must "think while moving".

The exploration-exploitation trade-off is at the heart of reinforcement learning (RL). However, most continuous control benchmarks used in recent RL research only require local exploration. This led to the development of algorithms that have basic exploration capabilities, and behave poorly in benchmarks that require more versatile exploration. For instance, as demonstrated in our empirical study, state-of-the-art RL algorithms such as DDPG and TD3 are unable to steer a point mass in even small 2D mazes. In this paper, we propose a new algorithm called "Plan, Backplay, Chain Skills" (PBCS) that combines motion planning and reinforcement learning to solve hard exploration environments. In a first phase, a motion planning algorithm is used to find a single good trajectory, then an RL algorithm is trained using a curriculum derived from the trajectory, by combining a variant of the Backplay algorithm and skill chaining. We show that this method outperforms state-of-the-art RL algorithms in 2D maze environments of various sizes, and is able to improve on the trajectory obtained by the motion planning phase.

In many applications, it is desirable to extract only the relevant information from complex input data, which involves making a decision about which input features are relevant. The information bottleneck method formalizes this as an information-theoretic optimization problem by maintaining an optimal tradeoff between compression (throwing away irrelevant input information), and predicting the target. In many problem settings, including the reinforcement learning problems we consider in this work, we might prefer to compress only part of the input. This is typically the case when we have a standard conditioning input, such as a state observation, and a "privileged" input, which might correspond to the goal of a task, the output of a costly planning algorithm, or communication with another agent. In such cases, we might prefer to compress the privileged input, either to achieve better generalization (e.g., with respect to goals) or to minimize access to costly information (e.g., in the case of communication). Practical implementations of the information bottleneck based on variational inference require access to the privileged input in order to compute the bottleneck variable, so although they perform compression, this compression operation itself needs unrestricted, lossless access. In this work, we propose the variational bandwidth bottleneck, which decides for each example on the estimated value of the privileged information before seeing it, i.e., only based on the standard input, and then accordingly chooses stochastically, whether to access the privileged input or not. We formulate a tractable approximation to this framework and demonstrate in a series of reinforcement learning experiments that it can improve generalization and reduce access to computationally costly information.

We investigate the evolution of the Q values for the implementation of Deep Q Learning (DQL) in the Stable Baselines library. Stable Baselines incorporates the latest Reinforcement Learning techniques and achieves superhuman performance in many game environments. However, for some simple non-game environments, the DQL in Stable Baselines can struggle to find the correct actions. In this paper we aim to understand the types of environment where this suboptimal behavior can happen, and also investigate the corresponding evolution of the Q values for individual states. We compare a smart TrafficLight environment (where performance is poor) with the AI Gym FrozenLake environment (where performance is perfect). We observe that DQL struggles with TrafficLight because actions are reversible and hence the Q values in a given state are closer than in FrozenLake. We then investigate the evolution of the Q values using a recent decomposition technique of Achiam et al.. We observe that for TrafficLight, the function approximation error and the complex relationships between the states lead to a situation where some Q values meander far from optimal.

Our effort is toward unifying GAN and DRL algorithms into a unifying AI model (AGI or general-purpose AI or artificial general intelligence which has general-purpose applications to: (A) offline learning (of stored data) like GAN in (un/semi-/fully-)SL setting such as big data analytics (mining) and visualization; (B) online learning (of real or simulated devices) like DRL in RL setting (with/out environment reward) such as (real or simulated) robotics and control; Our core proposal is adding an (generative/predictive) environment model to the actor-critic (model-free) architecture which results in a model-based actor-critic architecture with temporal-differencing (TD) error and an episodic memory. The proposed AI model is similar to (model-free) DDPG and therefore it's called model-based DDPG. To evaluate it, we compare it with (model-free) DDPG by applying them both to a variety (wide range) of independent simulated robotic and control task environments in OpenAI Gym and Unity Agents. Our initial limited experiments show that DRL and GAN in model-based actor-critic results in an incremental goal-driven intellignce required to solve each task with similar performance to (model-free) DDPG. Our future focus is to investigate the proposed AI model potential to: (A) unify DRL field inside AI by producing competitive performance compared to the best of model-based (PlaNet) and model-free (D4PG) approaches; (B) bridge the gap between AI and robotics communities by solving the important problem of reward engineering with learning the reward function by demonstration;

A research team working at Google's Robotics division and the Georgia Institute of Technology has figured out how to let four-legged robots learn to walk without needing any help from humans. As MIT Technology Review reports, the researchers implemented a deep reinforcement learning framework, which combines a multi-tasking learning procedure, an automatic reset controller, and a framework that's safety-constrained. A robot can train using the framework for 80 minutes at a time to gain experience without human interaction. It learns multiple directions of travel at the same time, allowing it to use a restricted training space effectively (and without ever getting stuck at the edges). At first the robot learns forward and backward motion on a flat surface, then on a soft mattress, and finally on a doormat with crevices. It's also then possible to autonomously teach the ability to turn left and right across the three different surface types.

Papers published at the Neural Information Processing Systems Conference.

Papers published at the Neural Information Processing Systems Conference.