Mathematician Richard E. Bellman, whose Bellman equation also known as a …

In this paper, we introduce ControlVAE, a novel model-based framework for learning generative motion control policies based on variational autoencoders (VAE). Our framework can learn a rich and flexible latent representation of skills and a skill-conditioned generative control policy from a diverse set of unorganized motion sequences, which enables the generation of realistic human behaviors by sampling in the latent space and allows high-level control policies to reuse the learned skills to accomplish a variety of downstream tasks. In the training of ControlVAE, we employ a learnable world model to realize direct supervision of the latent space and the control policy. This world model effectively captures the unknown dynamics of the simulation system, enabling efficient model-based learning of high-level downstream tasks. We also learn a state-conditional prior distribution in the VAE-based generative control policy, which generates a skill embedding that outperforms the non-conditional priors in downstream tasks. We demonstrate the effectiveness of ControlVAE using a diverse set of tasks, which allows realistic and interactive control of the simulated characters.

Monitoring and patrolling large water resources is a major challenge for conservation. The problem of acquiring data of an underlying environment that usually changes within time involves a proper formulation of the information. The use of Autonomous Surface Vehicles equipped with water quality sensor modules can serve as an early-warning system agents for contamination peak-detection, algae blooms monitoring, or oil-spill scenarios. In addition to information gathering, the vehicle must plan routes that are free of obstacles on non-convex maps. This work proposes a framework to obtain a collision-free policy that addresses the patrolling task for static and dynamic scenarios. Using information gain as a measure of the uncertainty reduction over data, it is proposed a Deep Q-Learning algorithm improved by a Q-Censoring mechanism for model-based obstacle avoidance. The obtained results demonstrate the usefulness of the proposed algorithm for water resource monitoring for static and dynamic scenarios. Simulations showed the use of noise-networks are a good choice for enhanced exploration, with 3 times less redundancy in the paths. Previous coverage strategies are also outperformed both in the accuracy of the obtained contamination model by a 13% on average and by a 37% in the detection of dangerous contamination peaks. Finally, these results indicate the appropriateness of the proposed framework for monitoring scenarios with autonomous vehicles.

There has been significant progress in developing reinforcement learning (RL) training systems. Past works such as IMPALA, Apex, Seed RL, Sample Factory, and others, aim to improve the system's overall throughput. In this paper, we aim to address a common bottleneck in the RL training system, i.e., parallel environment execution, which is often the slowest part of the whole system but receives little attention. With a curated design for paralleling RL environments, we have improved the RL environment simulation speed across different hardware setups, ranging from a laptop and a modest workstation, to a high-end machine such as NVIDIA DGX-A100. On a high-end machine, EnvPool achieves one million frames per second for the environment execution on Atari environments and three million frames per second on MuJoCo environments. When running EnvPool on a laptop, the speed is 2.8x that of the Python subprocess. Moreover, great compatibility with existing RL training libraries has been demonstrated in the open-sourced community, including CleanRL, rl_games, DeepMind Acme, etc. Finally, EnvPool allows researchers to iterate their ideas at a much faster pace and has great potential to become the de facto RL environment execution engine. Example runs show that it only takes five minutes to train agents to play Atari Pong and MuJoCo Ant on a laptop. EnvPool is open-sourced at https://github.com/sail-sg/envpool.

A good state representation is crucial to solving complicated reinforcement learning (RL) challenges. Many recent works focus on designing auxiliary losses for learning informative representations. Unfortunately, these handcrafted objectives rely heavily on expert knowledge and may be sub-optimal. In this paper, we propose a principled and universal method for learning better representations with auxiliary loss functions, named Automated Auxiliary Loss Search (A2LS), which automatically searches for top-performing auxiliary loss functions for RL. Specifically, based on the collected trajectory data, we define a general auxiliary loss space of size $7.5 \times 10^{20}$ and explore the space with an efficient evolutionary search strategy. Empirical results show that the discovered auxiliary loss (namely, A2-winner) significantly improves the performance on both high-dimensional (image) and low-dimensional (vector) unseen tasks with much higher efficiency, showing promising generalization ability to different settings and even different benchmark domains. We conduct a statistical analysis to reveal the relations between patterns of auxiliary losses and RL performance.

Being able to provide counterfactual interventions - sequences of actions we would have had to take for a desirable outcome to happen - is essential to explain how to change an unfavourable decision by a black-box machine learning model (e.g., being denied a loan request). Existing solutions have mainly focused on generating feasible interventions without providing explanations on their rationale. Moreover, they need to solve a separate optimization problem for each user. In this paper, we take a different approach and learn a program that outputs a sequence of explainable counterfactual actions given a user description and a causal graph. We leverage program synthesis techniques, reinforcement learning coupled with Monte Carlo Tree Search for efficient exploration, and rule learning to extract explanations for each recommended action. An experimental evaluation on synthetic and real-world datasets shows how our approach generates effective interventions by making orders of magnitude fewer queries to the black-box classifier with respect to existing solutions, with the additional benefit of complementing them with interpretable explanations.

This paper proposes, implements, and evaluates a reinforcement learning (RL)-based computational framework for automatic mesh generation. Mesh generation plays a fundamental role in numerical simulations in the area of computer aided design and engineering (CAD/E). It is identified as one of the critical issues in the NASA CFD Vision 2030 Study. Existing mesh generation methods suffer from high computational complexity, low mesh quality in complex geometries, and speed limitations. These methods and tools, including commercial software packages, are typically semiautomatic and they need inputs or help from human experts. By formulating the mesh generation as a Markov decision process (MDP) problem, we are able to use a state-of-the-art reinforcement learning (RL) algorithm called "soft actor-critic" to automatically learn from trials the policy of actions for mesh generation. The implementation of this RL algorithm for mesh generation allows us to build a fully automatic mesh generation system without human intervention and any extra clean-up operations, which fills the gap in the existing mesh generation tools. In the experiments to compare with two representative commercial software packages, our system demonstrates promising performance with respect to scalability, generalizability, and effectiveness.

Design time uncertainty poses an important challenge when developing a self-adaptive system. As an example, defining how the system should adapt when facing a new environment state, requires understanding the precise effect of an adaptation, which may not be known at design time. Online reinforcement learning, i.e., employing reinforcement learning (RL) at runtime, is an emerging approach to realizing self-adaptive systems in the presence of design time uncertainty. By using Online RL, the self-adaptive system can learn from actual operational data and leverage feedback only available at runtime. Recently, Deep RL is gaining interest. Deep RL represents learned knowledge as a neural network whereby it can generalize over unseen inputs, as well as handle continuous environment states and adaptation actions. A fundamental problem of Deep RL is that learned knowledge is not explicitly represented. For a human, it is practically impossible to relate the parametrization of the neural network to concrete RL decisions and thus Deep RL essentially appears as a black box. Yet, understanding the decisions made by Deep RL is key to (1) increasing trust, and (2) facilitating debugging. Such debugging is especially relevant for self-adaptive systems, because the reward function, which quantifies the feedback to the RL algorithm, must be defined by developers. The reward function must be explicitly defined by developers, thus introducing a potential for human error. To explain Deep RL for self-adaptive systems, we enhance and combine two existing explainable RL techniques from the machine learning literature. The combined technique, XRL-DINE, overcomes the respective limitations of the individual techniques. We present a proof-of-concept implementation of XRL-DINE, as well as qualitative and quantitative results of applying XRL-DINE to a self-adaptive system exemplar.

In risk-averse reinforcement learning (RL), the goal is to optimize some risk measure of the returns. A risk measure often focuses on the worst returns out of the agent's experience. As a result, standard methods for risk-averse RL often ignore high-return strategies. We prove that under certain conditions this inevitably leads to a local-optimum barrier, and propose a soft risk mechanism to bypass it. We also devise a novel Cross Entropy module for risk sampling, which (1) preserves risk aversion despite the soft risk; (2) independently improves sample efficiency. By separating the risk aversion of the sampler and the optimizer, we can sample episodes with poor conditions, yet optimize with respect to successful strategies. We combine these two concepts in CeSoR - Cross-entropy Soft-Risk optimization algorithm - which can be applied on top of any risk-averse policy gradient (PG) method. We demonstrate improved risk aversion in maze navigation, autonomous driving, and resource allocation benchmarks, including in scenarios where standard risk-averse PG completely fails.

Safe exploration is a challenging and important problem in model-free reinforcement learning (RL). Often the safety cost is sparse and unknown, which unavoidably leads to constraint violations -- a phenomenon ideally to be avoided in safety-critical applications. We tackle this problem by augmenting the state-space with a safety state, which is nonnegative if and only if the constraint is satisfied. The value of this state also serves as a distance toward constraint violation, while its initial value indicates the available safety budget. This idea allows us to derive policies for scheduling the safety budget during training. We call our approach Simmer (Safe policy IMproveMEnt for RL) to reflect the careful nature of these schedules. We apply this idea to two safe RL problems: RL with constraints imposed on an average cost, and RL with constraints imposed on a cost with probability one. Our experiments suggest that "simmering, a safe algorithm can improve safety during training for both settings. We further show that Simmer can stabilize training and improve the performance of safe RL with average constraints.