Zero-sum Markov Games (MGs) has been an efficient framework for multi-agent systems and robust control, wherein a minimax problem is constructed to solve the equilibrium policies. At present, this formulation is well studied under tabular settings wherein the maximum operator is primarily and exactly solved to calculate the worst-case value function. However, it is non-trivial to extend such methods to handle complex tasks, as finding the maximum over large-scale action spaces is usually cumbersome. In this paper, we propose the smoothing policy iteration (SPI) algorithm to solve the zero-sum MGs approximately, where the maximum operator is replaced by the weighted LogSumExp (WLSE) function to obtain the nearly optimal equilibrium policies. Specially, the adversarial policy is served as the weight function to enable an efficient sampling over action spaces.We also prove the convergence of SPI and analyze its approximation error in $\infty -$norm based on the contraction mapping theorem. Besides, we propose a model-based algorithm called Smooth adversarial Actor-critic (SaAC) by extending SPI with the function approximations. The target value related to WLSE function is evaluated by the sampled trajectories and then mean square error is constructed to optimize the value function, and the gradient-ascent-descent methods are adopted to optimize the protagonist and adversarial policies jointly. In addition, we incorporate the reparameterization technique in model-based gradient back-propagation to prevent the gradient vanishing due to sampling from the stochastic policies. We verify our algorithm in both tabular and function approximation settings. Results show that SPI can approximate the worst-case value function with a high accuracy and SaAC can stabilize the training process and improve the adversarial robustness in a large margin.

We present an approach for safe trajectory planning, where a strategic task related to autonomous racing is learned sample-efficient within a simulation environment. A high-level policy, represented as a neural network, outputs a reward specification that is used within the cost function of a parametric nonlinear model predictive controller (NMPC). By including constraints and vehicle kinematics in the NLP, we are able to guarantee safe and feasible trajectories related to the used model. Compared to classical reinforcement learning (RL), our approach restricts the exploration to safe trajectories, starts with a good prior performance and yields full trajectories that can be passed to a tracking lowest-level controller. We do not address the lowest-level controller in this work and assume perfect tracking of feasible trajectories. We show the superior performance of our algorithm on simulated racing tasks that include high-level decision making. The vehicle learns to efficiently overtake slower vehicles and to avoid getting overtaken by blocking faster vehicles.

Secondly, the communication improves its policy iteratively by learning from observations delay can interfere with the cooperation between agents to achieve a given goal. RL, with a single agent to decide by introducing delays in action-making (Chen et al. 2021) the behavior of all entities, faces various challenges, such and uncertainty on the arrival time of information. Previous as scalability (Yan et al. 2021) and privacy issues (Yuan, work (Kim et al. 2019) prevents endless waiting by setting Chung, and Fu 2022). To this end, the extension from singleagent a predefined and constant bound for the waiting time, but it RL to multi-agent RL (MARL) (Hernandez-Leal, Kartal, may restrain potential cooperation if it is set too short and and Taylor 2019) is favorable. MARL (Hernandez-Leal, conversely may cause meaningless waiting. Therefore, such Kartal, and Taylor 2019) has been widely used in various a constant timer is inflexible and cannot be adapted to the tasks, such as real-time resource allocation (Yuan et al. dynamics in the communication networks.

Meshing is a critical, but user-intensive process necessary for stable and accurate simulations in computational fluid dynamics (CFD). Mesh generation is often a bottleneck in CFD pipelines. Adaptive meshing techniques allow the mesh to be updated automatically to produce an accurate solution for the problem at hand. Existing classical techniques for adaptive meshing require either additional functionality out of solvers, many training simulations, or both. Current machine learning techniques often require substantial computational cost for training data generation, and are restricted in scope to the training data flow regime. MeshDQN is developed as a general purpose deep reinforcement learning framework to iteratively coarsen meshes while preserving target property calculation. A graph neural network based deep Q network is used to select mesh vertices for removal and solution interpolation is used to bypass expensive simulations at each step in the improvement process. MeshDQN requires a single simulation prior to mesh coarsening, while making no assumptions about flow regime, mesh type, or solver, only requiring the ability to modify meshes directly in a CFD pipeline. MeshDQN successfully improves meshes for two 2D airfoils.

ML-based motion planning is a promising approach to produce agents that exhibit complex behaviors, and automatically adapt to novel environments. In the context of autonomous driving, it is common to treat all available training data equally. However, this approach produces agents that do not perform robustly in safety-critical settings, an issue that cannot be addressed by simply adding more data to the training set - we show that an agent trained using only a 10% subset of the data performs just as well as an agent trained on the entire dataset. We present a method to predict the inherent difficulty of a driving situation given data collected from a fleet of autonomous vehicles deployed on public roads. We then demonstrate that this difficulty score can be used in a zero-shot transfer to generate curricula for an imitation-learning based planning agent. Compared to training on the entire unbiased training dataset, we show that prioritizing difficult driving scenarios both reduces collisions by 15% and increases route adherence by 14% in closed-loop evaluation, all while using only 10% of the training data.

Recent advances in Federated Learning (FL) have paved the way towards the design of novel strategies for solving multiple learning tasks simultaneously, by leveraging cooperation among networked devices. Multi-Task Learning (MTL) exploits relevant commonalities across tasks to improve efficiency compared with traditional transfer learning approaches. By learning multiple tasks jointly, significant reduction in terms of energy footprints can be obtained. This article provides a first look into the energy costs of MTL processes driven by the Model-Agnostic Meta-Learning (MAML) paradigm and implemented in distributed wireless networks. The paper targets a clustered multi-task network setup where autonomous agents learn different but related tasks. The MTL process is carried out in two stages: the optimization of a meta-model that can be quickly adapted to learn new tasks, and a task-specific model adaptation stage where the learned meta-model is transferred to agents and tailored for a specific task. This work analyzes the main factors that influence the MTL energy balance by considering a multi-task Reinforcement Learning (RL) setup in a robotized environment. Results show that the MAML method can reduce the energy bill by at least 2 times compared with traditional approaches without inductive transfer. Moreover, it is shown that the optimal energy balance in wireless networks depends on uplink/downlink and sidelink communication efficiencies.

Multi-agent reinforcement learning (MARL) suffers from the non-stationarity problem, which is the ever-changing targets at every iteration when multiple agents update their policies at the same time. Starting from first principle, in this paper, we manage to solve the non-stationarity problem by proposing bidirectional action-dependent Q-learning (ACE). Central to the development of ACE is the sequential decision-making process wherein only one agent is allowed to take action at one time. Within this process, each agent maximizes its value function given the actions taken by the preceding agents at the inference stage. In the learning phase, each agent minimizes the TD error that is dependent on how the subsequent agents have reacted to their chosen action. Given the design of bidirectional dependency, ACE effectively turns a multiagent MDP into a single-agent MDP. We implement the ACE framework by identifying the proper network representation to formulate the action dependency, so that the sequential decision process is computed implicitly in one forward pass. To validate ACE, we compare it with strong baselines on two MARL benchmarks. Empirical experiments demonstrate that ACE outperforms the state-of-the-art algorithms on Google Research Football and StarCraft Multi-Agent Challenge by a large margin. In particular, on SMAC tasks, ACE achieves 100% success rate on almost all the hard and super-hard maps. We further study extensive research problems regarding ACE, including extension, generalization, and practicability. Code is made available to facilitate further research.

In reinforcement learning, different reward functions can be equivalent in terms of the optimal policies they induce. A particularly well-known and important example is potential shaping, a class of functions that can be added to any reward function without changing the optimal policy set under arbitrary transition dynamics. Potential shaping is conceptually similar to potentials, conservative vector fields and gauge transformations in math and physics, but this connection has not previously been formally explored. We develop a formalism for discrete calculus on graphs that abstract a Markov Decision Process, and show how potential shaping can be formally interpreted as a gradient within this framework. This allows us to strengthen results from Ng et al. (1999) describing conditions under which potential shaping is the only additive reward transformation to always preserve optimal policies. As an additional application of our formalism, we define a rule for picking a single unique reward function from each potential shaping equivalence class.

Robots need to be able to adapt to unexpected changes in the environment such that they can autonomously succeed in their tasks. However, hand-designing feedback models for adaptation is tedious, if at all possible, making data-driven methods a promising alternative. In this paper we introduce a full framework for learning feedback models for reactive motion planning. Our pipeline starts by segmenting demonstrations of a complete task into motion primitives via a semi-automated segmentation algorithm. Then, given additional demonstrations of successful adaptation behaviors, we learn initial feedback models through learning from demonstrations. In the final phase, a sample-efficient reinforcement learning algorithm fine-tunes these feedback models for novel task settings through few real system interactions. We evaluate our approach on a real anthropomorphic robot in learning a tactile feedback task.

Deep reinforcement learning algorithms have succeeded in several challenging domains. Classic Online RL job schedulers can learn efficient scheduling strategies but often takes thousands of timesteps to explore the environment and adapt from a randomly initialized DNN policy. Existing RL schedulers overlook the importance of learning from historical data and improving upon custom heuristic policies. Offline reinforcement learning presents the prospect of policy optimization from pre-recorded datasets without online environment interaction. Following the recent success of data-driven learning, we explore two RL methods: 1) Behaviour Cloning and 2) Offline RL, which aim to learn policies from logged data without interacting with the environment. These methods address the challenges concerning the cost of data collection and safety, particularly pertinent to real-world applications of RL. Although the data-driven RL methods generate good results, we show that the performance is highly dependent on the quality of the historical datasets. Finally, we demonstrate that by effectively incorporating prior expert demonstrations to pre-train the agent, we short-circuit the random exploration phase to learn a reasonable policy with online training. We utilize Offline RL as a launchpad to learn effective scheduling policies from prior experience collected using Oracle or heuristic policies. Such a framework is effective for pre-training from historical datasets and well suited to continuous improvement with online data collection.