Designing controllers to achieve natural motor capabilities for multi-joint robots is a significant challenge. However, animals in nature are naturally with basic motor abilities and can master various complex motor skills through acquired learning. On the basis of analyzing the mechanism of the central motor system in mammals, we propose a novel pre-training reinforcement learning algorithm that enables robots to learn rich motor skills and apply them to complex task environments without relying on external data. We first design a skill based network similar to the cerebellum by utilizing the selection mechanism of voluntary movements in the basal ganglia and the basic motor regulation ability of the cerebellum. Subsequently, by imitating the structure of advanced centers in the central motor system, we propose a high-level policy to generate different skill combinations, thereby enabling the robot to acquire natural motor abilities. We conduct experiments on 4 types of robots and 22 task environments, and the results show that the proposed method can enable different types of robots to achieve flexible motor skills. Overall, our research provides a promising framework for the design of neural network motor controllers.

Current evaluation protocols predominantly assess physical reasoning in stationary scenes, creating a gap in evaluating agents' abilities to interact with dynamic events. While contemporary methods allow agents to modify initial scene configurations and observe consequences, they lack the capability to interact with events in real time. To address this, we introduce I-PHYRE, a framework that challenges agents to simultaneously exhibit intuitive physical reasoning, multi-step planning, and in-situ intervention. Here, intuitive physical reasoning refers to a quick, approximate understanding of physics to address complex problems; multi-step denotes the need for extensive sequence planning in I-PHYRE, considering each intervention can significantly alter subsequent choices; and in-situ implies the necessity for timely object manipulation within a scene, where minor timing deviations can result in task failure. We formulate four game splits to scrutinize agents' learning and generalization of essential principles of interactive physical reasoning, fostering learning through interaction with representative scenarios. Our exploration involves three planning strategies and examines several supervised and reinforcement agents' zero-shot generalization proficiency on I-PHYRE. The outcomes highlight a notable gap between existing learning algorithms and human performance, emphasizing the imperative for more research in enhancing agents with interactive physical reasoning capabilities. The environment and baselines will be made publicly available.

In response to the growing uptake of distributed energy resources (DERs), community batteries have emerged as a promising solution to support renewable energy integration, reduce peak load, and enhance grid reliability. This paper presents a deep reinforcement learning (RL) strategy, centered around the soft actor-critic (SAC) algorithm, to schedule a community battery system in the presence of uncertainties, such as solar photovoltaic (PV) generation, local demand, and real-time energy prices. We position the community battery to play a versatile role, in integrating local PV energy, reducing peak load, and exploiting energy price fluctuations for arbitrage, thereby minimizing the system cost. To improve exploration and convergence during RL training, we utilize the noisy network technique. This paper conducts a comparative study of different RL algorithms, including proximal policy optimization (PPO) and deep deterministic policy gradient (DDPG) algorithms, to evaluate their effectiveness in the community battery scheduling problem. The results demonstrate the potential of RL in addressing community battery scheduling challenges and show that the SAC algorithm achieves the best performance compared to RL and optimization benchmarks.

Unmanned aerial vehicle (UAV) is a promising technology for last-mile cargo delivery. However, the limited on-board battery capacity, cellular unreliability, and frequent handoffs in the airspace are the main obstacles to unleash its full potential. Given that existing cellular networks were primarily designed to service ground users, re-utilizing the same architecture for highly mobile aerial users, e.g., cargo-UAVs, is deemed challenging. Indeed, to ensure a safe delivery using cargo-UAVs, it is crucial to utilize the available energy efficiently, while guaranteeing reliable connectivity for command-and-control and avoiding frequent handoff. To achieve this goal, we propose a novel approach for joint cargo-UAV trajectory planning and cell association. Specifically, we formulate the cargo-UAV mission as a multi-objective problem aiming to 1) minimize energy consumption, 2) reduce handoff events, and 3) guarantee cellular reliability along the trajectory. We leverage reinforcement learning (RL) to jointly optimize the cargo-UAV's trajectory and cell association. Simulation results demonstrate a performance improvement of our proposed method, in terms of handoffs, disconnectivity, and energy consumption, compared to benchmarks.

Offline reinforcement learning algorithms often require careful hyperparameter tuning. Consequently, before deployment, we need to select amongst a set of candidate policies. As yet, however, there is little understanding about the fundamental limits of this offline policy selection (OPS) problem. In this work we aim to provide clarity on when sample efficient OPS is possible, primarily by connecting OPS to off-policy policy evaluation (OPE) and Bellman error (BE) estimation. We first show a hardness result, that in the worst case, OPS is just as hard as OPE, by proving a reduction of OPE to OPS. As a result, no OPS method can be more sample efficient than OPE in the worst case. We then propose a BE method for OPS, called Identifiable BE Selection (IBES), that has a straightforward method for selecting its own hyperparameters. We highlight that using IBES for OPS generally has more requirements than OPE methods, but if satisfied, can be more sample efficient. We conclude with an empirical study comparing OPE and IBES, and by showing the difficulty of OPS on an offline Atari benchmark dataset.

Video games have served as useful benchmarks for the decision making community, but going beyond Atari games towards training agents in modern games has been prohibitively expensive for the vast majority of the research community. Recent progress in the research, development and open release of large vision models has the potential to amortize some of these costs across the community. However, it is currently unclear which of these models have learnt representations that retain information critical for sequential decision making. Towards enabling wider participation in the research of gameplaying agents in modern games, we present a systematic study of imitation learning with publicly available visual encoders compared to the typical, task-specific, end-to-end training approach in Minecraft, Minecraft Dungeons and Counter-Strike: Global Offensive. Figure 1: Representative screenshots of all games studied in this paper. However, video games do not only serve as benchmarks but also represent a vast entertainment industry where AI agents may eventually have applications in games development, including game testing or game design (Jacob et al., 2020; Gillberg et al., 2023). In the past, video game research often necessitated close integration with the games themselves to obtain game-specific information and establish a scalable interface for training agents. Work was conducted during an internship at Microsoft Research. To eliminate integration costs during training, we use behavior cloning to train agents entirely offline, utilising previously collected human gameplay data. Although prior research has explored encoding images into lower-dimensional representations for behavior cloning, these studies primarily targeted robotics applications (Nair et al., 2022), where images often resemble real-world scenes. Inspired by the challenges and potential applications in video games, we investigate the following research question: How can images be encoded for data-efficient imitation learning in modern video games? Towards our guiding research question, we compare both end-to-end trained visual encoders and pre-trained visual encoders in three modern video games: Minecraft, Minecraft Dungeons and Counter-Strike: Global Offensive (CS:GO).

We introduce Parameterized Environment Response Model (PERM), a method for training both Reinforcement Learning (RL) Agents and human learners in parameterized environments by directly modeling difficulty and ability. Inspired by Item Response Theory (IRT), PERM aligns environment difficulty with individual ability, creating a Zone of Proximal Development-based curriculum. Remarkably, PERM operates without real-time RL updates and allows for offline training, ensuring its adaptability across diverse students. We present a two-stage training process that capitalizes on PERM's adaptability, and demonstrate its effectiveness in training RL agents and humans in an empirical study. Figure 1: Overview of the proposed 2-stage process. In Stage 1, the IRT-based Parameterized Environment Response Model (PERM) observes a Reinforcement Learning (RL) Agent as it trains in a given environment with randomized levels. During this stage, PERM learns to accurately infer both student ability and level difficulty. In Stage 2, once trained, PERM is deployed to train both artificial and human students. It achieves this by inferring their current ability and providing suitable training levels within the same domain.

A central challenge of the clean energy transition is the development of catalysts for low-emissions technologies. Recent advances in Machine Learning for quantum chemistry drastically accelerate the computation of catalytic activity descriptors such as adsorption energies. Here we introduce AdsorbRL, a Deep Reinforcement Learning agent aiming to identify potential catalysts given a multi-objective binding energy target, trained using offline learning on the Open Catalyst 2020 and Materials Project data sets. We experiment with Deep Q-Network agents to traverse the space of all ~160,000 possible unary, binary and ternary compounds of 55 chemical elements, with very sparse rewards based on adsorption energy known for only between 2,000 and 3,000 catalysts per adsorbate. To constrain the actions space, we introduce Random Edge Traversal and train a single-objective DQN agent on the known states subgraph, which we find strengthens target binding energy by an average of 4.1 eV. We extend this approach to multi-objective, goal-conditioned learning, and train a DQN agent to identify materials with the highest (respectively lowest) adsorption energies for multiple simultaneous target adsorbates. We experiment with Objective Sub-Sampling, a novel training scheme aimed at encouraging exploration in the multi-objective setup, and demonstrate simultaneous adsorption energy improvement across all target adsorbates, by an average of 0.8 eV. Overall, our results suggest strong potential for Deep Reinforcement Learning applied to the inverse catalysts design problem.

Unlike most reinforcement learning agents which require an unrealistic amount of environment interactions to learn a new behaviour, humans excel at learning quickly by merely observing and imitating others. This ability highly depends on the fact that humans have a model of their own embodiment that allows them to infer the most likely actions that led to the observed behaviour. In this paper, we propose Action Inference by Maximising Evidence (AIME) to replicate this behaviour using world models. AIME consists of two distinct phases. In the first phase, the agent learns a world model from its past experience to understand its own body by maximising the ELBO. While in the second phase, the agent is given some observation-only demonstrations of an expert performing a novel task and tries to imitate the expert's behaviour. AIME achieves this by defining a policy as an inference model and maximising the evidence of the demonstration under the policy and world model. Our method is "zero-shot" in the sense that it does not require further training for the world model or online interactions with the environment after given the demonstration. We empirically validate the zero-shot imitation performance of our method on the Walker and Cheetah embodiment of the DeepMind Control Suite and find it outperforms the state-of-the-art baselines. Code is available at: https://github.com/argmax-ai/aime.

Increasing interest in ensuring safety of next-generation Artificial Intelligence (AI) systems calls for novel approaches to embedding morality into autonomous agents. Traditionally, this has been done by imposing explicit top-down rules or hard constraints on systems, for example by filtering system outputs through pre-defined ethical rules. Recently, instead, entirely bottom-up methods for learning implicit preferences from human behavior have become increasingly popular, such as those for training and fine-tuning Large Language Models. In this paper, we provide a systematization of existing approaches to the problem of introducing morality in machines - modeled as a continuum, and argue that the majority of popular techniques lie at the extremes - either being fully hard-coded, or entirely learned, where no explicit statement of any moral principle is required. Given the relative strengths and weaknesses of each type of methodology, we argue that more hybrid solutions are needed to create adaptable and robust, yet more controllable and interpretable agents. In particular, we present three case studies of recent works which use learning from experience (i.e., Reinforcement Learning) to explicitly provide moral principles to learning agents - either as intrinsic rewards, moral logical constraints or textual principles for language models. For example, using intrinsic rewards in Social Dilemma games, we demonstrate how it is possible to represent classical moral frameworks for agents. We also present an overview of the existing work in this area in order to provide empirical evidence for the potential of this hybrid approach. We then discuss strategies for evaluating the effectiveness of moral learning agents. Finally, we present open research questions and implications for the future of AI safety and ethics which are emerging from this framework.