General Value Function (GVF) is a powerful tool to represent both the {\em predictive} and {\em retrospective} knowledge in reinforcement learning (RL). In practice, often multiple interrelated GVFs need to be evaluated jointly with pre-collected off-policy samples. In the literature, the gradient temporal difference (GTD) learning method has been adopted to evaluate GVFs in the off-policy setting, but such an approach may suffer from a large estimation error even if the function approximation class is sufficiently expressive. Moreover, none of the previous work have formally established the convergence guarantee to the ground truth GVFs under the function approximation settings. In this paper, we address both issues through the lens of a class of GVFs with causal filtering, which cover a wide range of RL applications such as reward variance, value gradient, cost in anomaly detection, stationary distribution gradient, etc. We propose a new algorithm called GenTD for off-policy GVFs evaluation and show that GenTD learns multiple interrelated multi-dimensional GVFs as efficiently as a single canonical scalar value function. We further show that unlike GTD, the learned GVFs by GenTD are guaranteed to converge to the ground truth GVFs as long as the function approximation power is sufficiently large. To our best knowledge, GenTD is the first off-policy GVF evaluation algorithm that has global optimality guarantee.

In Meta-Reinforcement Learning (meta-RL) an agent is trained on a set of tasks to prepare for and learn faster in new, unseen, but related tasks. The training tasks are usually hand-crafted to be representative of the expected distribution of test tasks and hence all used in training. We show that given a set of training tasks, learning can be both faster and more effective (leading to better performance in the test tasks), if the training tasks are appropriately selected. We propose a task selection algorithm, Information-Theoretic Task Selection (ITTS), based on information theory, which optimizes the set of tasks used for training in meta-RL, irrespectively of how they are generated. The algorithm establishes which training tasks are both sufficiently relevant for the test tasks, and different enough from one another. We reproduce different meta-RL experiments from the literature and show that ITTS improves the final performance in all of them.

Eluder dimension and information gain are two widely used methods of complexity measures in bandit and reinforcement learning. Eluder dimension was originally proposed as a general complexity measure of function classes, but the common examples of where it is known to be small are function spaces (vector spaces). In these cases, the primary tool to upper bound the eluder dimension is the elliptic potential lemma. Interestingly, the elliptic potential lemma also features prominently in the analysis of linear bandits/reinforcement learning and their nonparametric generalization, the information gain. We show that this is not a coincidence -- eluder dimension and information gain are equivalent in a precise sense for reproducing kernel Hilbert spaces.

Model-based deep reinforcement learning has achieved success in various domains that require high sample efficiencies, such as Go and robotics. However, there are some remaining issues, such as planning efficient explorations to learn more accurate dynamic models, evaluating the uncertainty of the learned models, and more rational utilization of models. To mitigate these issues, we present MEEE, a model-ensemble method that consists of optimistic exploration and weighted exploitation. During exploration, unlike prior methods directly selecting the optimal action that maximizes the expected accumulative return, our agent first generates a set of action candidates and then seeks out the optimal action that takes both expected return and future observation novelty into account. During exploitation, different discounted weights are assigned to imagined transition tuples according to their model uncertainty respectively, which will prevent model predictive error propagation in agent training. Experiments on several challenging continuous control benchmark tasks demonstrated that our approach outperforms other model-free and model-based state-of-the-art methods, especially in sample complexity.

Adaptive traffic signal control (ATSC) in urban traffic networks poses a challenging task due to the complicated dynamics arising in traffic systems. In recent years, several approaches based on multi-agent deep reinforcement learning (MARL) have been studied experimentally. These approaches propose distributed techniques in which each signalized intersection is seen as an agent in a stochastic game whose purpose is to optimize the flow of vehicles in its vicinity. In this setting, the systems evolves towards an equilibrium among the agents that shows beneficial for the whole traffic network. A recently developed multi-agent variant of the well-established advantage actor-critic (A2C) algorithm, called MA2C (multi-agent A2C) exploits the promising idea of some communication among the agents. In this view,the agents share their strategies with other neighbor agents, thereby stabilizing the learning process even when the agents grow in number and variety. We experimented MA2C in two traffic networks located in Bologna (Italy) and found that its action translates into a significant decrease of the amount of pollutants released into the environment.

This work focuses on learning useful and robust deep world models using multiple, possibly unreliable, sensors. We find that current methods do not sufficiently encourage a shared representation between modalities; this can cause poor performance on downstream tasks and over-reliance on specific sensors. As a solution, we contribute a new multi-modal deep latent state-space model, trained using a mutual information lower-bound. The key innovation is a specially-designed density ratio estimator that encourages consistency between the latent codes of each modality. We tasked our method to learn policies (in a self-supervised manner) on multi-modal Natural MuJoCo benchmarks and a challenging Table Wiping task. Experiments show our method significantly outperforms state-of-the-art deep reinforcement learning methods, particularly in the presence of missing observations.

Humans and other intelligent animals evolved highly sophisticated perception systems that combine multiple sensory modalities. On the other hand, state-of-the-art artificial agents rely mostly on visual inputs or structured low-dimensional observations provided by instrumented environments. Learning to act based on combined visual and auditory inputs is still a new topic of research that has not been explored beyond simple scenarios. To facilitate progress in this area we introduce a new version of VizDoom simulator to create a highly efficient learning environment that provides raw audio observations. We study the performance of different model architectures in a series of tasks that require the agent to recognize sounds and execute instructions given in natural language. Finally, we train our agent to play the full game of Doom and find that it can consistently defeat a traditional vision-based adversary. We are currently in the process of merging the augmented simulator with the main ViZDoom code repository. Video demonstrations and experiment code can be found at https://sites.google.com/view/sound-rl.

The last decade has seen a significant increase of interest in deep learning research, with many public successes that have demonstrated its potential. As such, these systems are now being incorporated into commercial products. With this comes an additional challenge: how can we build AI systems that solve tasks where there is not a crisp, well-defined specification? While multiple solutions have been proposed, in this competition we focus on one in particular: learning from human feedback. Rather than training AI systems using a predefined reward function or using a labeled dataset with a predefined set of categories, we instead train the AI system using a learning signal derived from some form of human feedback, which can evolve over time as the understanding of the task changes, or as the capabilities of the AI system improve. The MineRL BASALT competition aims to spur forward research on this important class of techniques. We design a suite of four tasks in Minecraft for which we expect it will be hard to write down hardcoded reward functions. These tasks are defined by a paragraph of natural language: for example, "create a waterfall and take a scenic picture of it", with additional clarifying details. Participants must train a separate agent for each task, using any method they want. Agents are then evaluated by humans who have read the task description. To help participants get started, we provide a dataset of human demonstrations on each of the four tasks, as well as an imitation learning baseline that leverages these demonstrations. Our hope is that this competition will improve our ability to build AI systems that do what their designers intend them to do, even when the intent cannot be easily formalized. Besides allowing AI to solve more tasks, this can also enable more effective regulation of AI systems, as well as making progress on the value alignment problem.

Ensemble and auxiliary tasks are both well known to improve the performance of machine learning models when data is limited. However, the interaction between these two methods is not well studied, particularly in the context of deep reinforcement learning. In this paper, we study the effects of ensemble and auxiliary tasks when combined with the deep Q-learning algorithm. We perform a case study on ATARI games under limited data constraint. Moreover, we derive a refined bias-variance-covariance decomposition to analyze the different ways of learning ensembles and using auxiliary tasks, and use the analysis to help provide some understanding of the case study. Our code is open source and available at https://github.com/NUS-LID/RENAULT.

ABSTRACT We explore the potential to control terrain vehicles using deep reinforcement in scenarios where human operators and traditional control methods are inadequate. This letter presents a controller that perceives, plans, and successfully controls a 16-tonne forestry vehicle with two frame articulation joints, six wheels, and their actively articulated suspensions to traverse rough terrain. The carefully shaped reward signal promotes safe, environmental, and efficient driving, which leads to the emergence of unprecedented driving skills. We test learned skills in a virtual environment, including terrains reconstructed from high-density laser scans of forest sites. The results confirm that deep reinforcement learning has the potential to enhance control of vehicles with complex dynamics and high-dimensional observation data compared to human operators or traditional control methods, especially in rough terrain. 1 INTRODUCTION Deep reinforcement learning has recently shown promise for locomotion tasks, but its usefulness to learn control of heavy vehicles in rough terrain is widely unknown. Conventionally, the design of rough terrain vehicles strives to promote high traversability and be easily operated by humans. The drivelines involve differentials and bogie suspension that provide ground compliance and reduces the many degrees of freedom, leaving only speed and heading for the operator to control. An attractive alternative is to use actively articulated suspensions and individual wheel control. These have the potential to reduce the energy consumption and ground damage, yet increase traversability and tip over stability [11, 6, 21, 10, 9].