Experience replay (ER) is a crucial component of many deep reinforcement learning (RL) systems. However, uniform sampling from an ER buffer can lead to slow convergence and unstable asymptotic behaviors. This paper introduces Stratified Sampling from Event Tables (SSET), which partitions an ER buffer into Event Tables, each capturing important subsequences of optimal behavior. We prove a theoretical advantage over the traditional monolithic buffer approach and combine SSET with an existing prioritized sampling strategy to further improve learning speed and stability. Empirical results in challenging MiniGrid domains, benchmark RL environments, and a high-fidelity car racing simulator demonstrate the advantages and versatility of SSET over existing ER buffer sampling approaches.

Partially-observable problems pose a trade-off between reducing costs and gathering information. They can be solved optimally by planning in belief space, but that is often prohibitively expensive. Model-predictive control (MPC) takes the alternative approach of using a state estimator to form a belief over the state, and then plan in state space. This ignores potential future observations during planning and, as a result, cannot actively increase or preserve the certainty of its own state estimate. We find a middle-ground between planning in belief space and completely ignoring its dynamics by only reasoning about its future accuracy. Our approach, filter-aware MPC, penalises the loss of information by what we call "trackability", the expected error of the state estimator. We show that model-based simulation allows condensing trackability into a neural network, which allows fast planning. In experiments involving visual navigation, realistic every-day environments and a two-link robot arm, we show that filter-aware MPC vastly improves regular MPC.

There has recently been an increased interest in reinforcement learning for nonlinear control problems. However standard reinforcement learning algorithms can often struggle even on seemingly simple set-point control problems. This paper argues that three ideas can improve reinforcement learning methods even for highly nonlinear set-point control problems: 1) Make use of a prior feedback controller to aid amplitude exploration. 2) Use integrated errors. 3) Train on model ensembles. Together these ideas lead to more efficient training, and a trained set-point controller that is more robust to modelling errors and thus can be directly deployed to real-world nonlinear systems. The claim is supported by experiments with a real-world nonlinear cascaded tank process and a simulated strongly nonlinear pH-control system.

While reinforcement learning has made great improvements, state-of-the-art algorithms can still struggle with seemingly simple set-point feedback control problems. One reason for this is that the learned controller may not be able to excite the system dynamics well enough initially, and therefore it can take a long time to get data that is informative enough to learn for good control. The paper contributes by augmentation of reinforcement learning with a simple guiding feedback controller, for example, a proportional controller. The key advantage in set point control is a much improved excitation that improves the convergence properties of the reinforcement learning controller significantly. This can be very important in real-world control where quick and accurate convergence is needed. The proposed method is evaluated with simulation and on a real-world double tank process with promising results.

Power grids, across the world, play an important societal and economical role by providing uninterrupted, reliable and transient-free power to several industries, businesses and household consumers. With the advent of renewable power resources and EVs resulting into uncertain generation and highly dynamic load demands, it has become ever so important to ensure robust operation of power networks through suitable management of transient stability issues and localize the events of blackouts. In the light of ever increasing stress on the modern grid infrastructure and the grid operators, this paper presents a reinforcement learning (RL) framework, PowRL, to mitigate the effects of unexpected network events, as well as reliably maintain electricity everywhere on the network at all times. The PowRL leverages a novel heuristic for overload management, along with the RL-guided decision making on optimal topology selection to ensure that the grid is operated safely and reliably (with no overloads). PowRL is benchmarked on a variety of competition datasets hosted by the L2RPN (Learning to Run a Power Network). Even with its reduced action space, PowRL tops the leaderboard in the L2RPN NeurIPS 2020 challenge (Robustness track) at an aggregate level, while also being the top performing agent in the L2RPN WCCI 2020 challenge. Moreover, detailed analysis depicts state-of-the-art performances by the PowRL agent in some of the test scenarios.

Questions remain on the robustness of data-driven learning methods when crossing the gap from simulation to reality. We utilize weight anchoring, a method known from continual learning, to cultivate and fixate desired behavior in Neural Networks. Weight anchoring may be used to find a solution to a learning problem that is nearby the solution of another learning problem. Thereby, learning can be carried out in optimal environments without neglecting or unlearning desired behavior. We demonstrate this approach on the example of learning mixed QoS-efficient discrete resource scheduling with infrequent priority messages. Results show that this method provides performance comparable to the state of the art of augmenting a simulation environment, alongside significantly increased robustness and steerability.

Many recent works have turned to multi-agent reinforcement learning (MARL) for adaptive traffic signal control to optimize the travel time of vehicles over large urban networks. However, achieving effective and scalable cooperation among junctions (agents) remains an open challenge, as existing methods often rely on extensive, non-generalizable reward shaping or on non-scalable centralized learning. To address these problems, we propose a new MARL method for traffic signal control, SocialLight, which learns cooperative traffic control policies by distributedly estimating the individual marginal contribution of agents on their local neighborhood. SocialLight relies on the Asynchronous Actor Critic (A3C) framework, and makes learning scalable by learning a locally-centralized critic conditioned over the states and actions of neighboring agents, used by agents to estimate individual contributions by counterfactual reasoning. We further introduce important modifications to the advantage calculation that help stabilize policy updates. These modifications decouple the impact of the neighbors' actions on the computed advantages, thereby reducing the variance in the gradient updates. We benchmark our trained network against state-of-the-art traffic signal control methods on standard benchmarks in two traffic simulators, SUMO and CityFlow. Our results show that SocialLight exhibits improved scalability to larger road networks and better performance across usual traffic metrics.

Picking cluttered general objects is a challenging task due to the complex geometries and various stacking configurations. Many prior works utilize pose estimation for picking, but pose estimation is difficult on cluttered objects. In this paper, we propose Cluttered Objects Descriptors (CODs), a dense cluttered objects descriptor that can represent rich object structures, and use the pre-trained CODs network along with its intermediate outputs to train a picking policy. Additionally, we train the policy with reinforcement learning, which enable the policy to learn picking without supervision. We conduct experiments to demonstrate that our CODs is able to consistently represent seen and unseen cluttered objects, which allowed for the picking policy to robustly pick cluttered general objects. The resulting policy can pick 96.69% of unseen objects in our experimental environment which is twice as cluttered as the training scenarios.

Can conversational videos captured from multiple egocentric viewpoints reveal the map of a scene in a cost-efficient way? We seek to answer this question by proposing a new problem: efficiently building the map of a previously unseen 3D environment by exploiting shared information in the egocentric audio-visual observations of participants in a natural conversation. Our hypothesis is that as multiple people ("egos") move in a scene and talk among themselves, they receive rich audio-visual cues that can help uncover the unseen areas of the scene. Given the high cost of continuously processing egocentric visual streams, we further explore how to actively coordinate the sampling of visual information, so as to minimize redundancy and reduce power use. To that end, we present an audio-visual deep reinforcement learning approach that works with our shared scene mapper to selectively turn on the camera to efficiently chart out the space. We evaluate the approach using a state-of-the-art audio-visual simulator for 3D scenes as well as real-world video. Our model outperforms previous state-of-the-art mapping methods, and achieves an excellent cost-accuracy tradeoff. Project: http://vision.cs.utexas.edu/projects/chat2map.

We propose a model-free reinforcement learning architecture, called distributed attentional actor architecture after conditional attention (DA6-X), to provide better interpretability of conditional coordinated behaviors. The underlying principle involves reusing the saliency vector, which represents the conditional states of the environment, such as the global position of agents. Hence, agents with DA6-X flexibility built into their policy exhibit superior performance by considering the additional information in the conditional states during the decision-making process. The effectiveness of the proposed method was experimentally evaluated by comparing it with conventional methods in an objects collection game. By visualizing the attention weights from DA6-X, we confirmed that agents successfully learn situation-dependent coordinated behaviors by correctly identifying various conditional states, leading to improved interpretability of agents along with superior performance.