Learning policies via preference-based reward learning is an increasingly popular method for customizing agent behavior, but has been shown anecdotally to be prone to spurious correlations and reward hacking behaviors. While much prior work focuses on causal confusion in reinforcement learning and behavioral cloning, we focus on a systematic study of causal confusion and reward misidentification when learning from preferences. In particular, we perform a series of sensitivity and ablation analyses on several benchmark domains where rewards learned from preferences achieve minimal test error but fail to generalize to out-of-distribution states -- resulting in poor policy performance when optimized. We find that the presence of non-causal distractor features, noise in the stated preferences, and partial state observability can all exacerbate reward misidentification. We also identify a set of methods with which to interpret misidentified learned rewards. In general, we observe that optimizing misidentified rewards drives the policy off the reward's training distribution, resulting in high predicted (learned) rewards but low true rewards. These findings illuminate the susceptibility of preference learning to reward misidentification and causal confusion -- failure to consider even one of many factors can result in unexpected, undesirable behavior.

Traffic signal control has the potential to reduce congestion in dynamic networks. Recent studies show that traffic signal control with reinforcement learning (RL) methods can significantly reduce the average waiting time. However, a shortcoming of existing methods is that they require model retraining for new intersections with different structures. In this paper, we propose a novel reinforcement learning approach with augmented data (ADLight) to train a universal model for intersections with different structures. We propose a new agent design incorporating features on movements and actions with set current phase duration to allow the generalized model to have the same structure for different intersections. A new data augmentation method named \textit{movement shuffle} is developed to improve the generalization performance. We also test the universal model with new intersections in Simulation of Urban MObility (SUMO). The results show that the performance of our approach is close to the models trained in a single environment directly (only a 5% loss of average waiting time), and we can reduce more than 80% of training time, which saves a lot of computational resources in scalable operations of traffic lights.

Stochastic optimal control and games have found a wide range of applications, from finance and economics to social sciences, robotics and energy management. Many real-world applications involve complex models which have driven the development of sophisticated numerical methods. Recently, computational methods based on machine learning have been developed for stochastic control problems and games. We review such methods, with a focus on deep learning algorithms that have unlocked the possibility to solve such problems even when the dimension is high or when the structure is very complex, beyond what is feasible with traditional numerical methods. Here, we consider mostly the continuous time and continuous space setting. Many of the new approaches build on recent neural-network based methods for high-dimensional partial differential equations or backward stochastic differential equations, or on model-free reinforcement learning for Markov decision processes that have led to breakthrough results. In this paper we provide an introduction to these methods and summarize state-of-the-art works on machine learning for stochastic control and games.

The flock-guidance problem enjoys a challenging structure where multiple optimization objectives are solved simultaneously. This usually necessitates different control approaches to tackle various objectives, such as guidance, collision avoidance, and cohesion. The guidance schemes, in particular, have long suffered from complex tracking-error dynamics. Furthermore, techniques that are based on linear feedback strategies obtained at equilibrium conditions either may not hold or degrade when applied to uncertain dynamic environments. Pre-tuned fuzzy inference architectures lack robustness under such unmodeled conditions. This work introduces an adaptive distributed technique for the autonomous control of flock systems. Its relatively flexible structure is based on online fuzzy reinforcement learning schemes which simultaneously target a number of objectives; namely, following a leader, avoiding collision, and reaching a flock velocity consensus. In addition to its resilience in the face of dynamic disturbances, the algorithm does not require more than the agent position as a feedback signal. The effectiveness of the proposed method is validated with two simulation scenarios and benchmarked against a similar technique from the literature.

Real-time Digital Twinning of physical world scenes onto the Metaverse is necessary for a myriad of applications such as augmented-reality (AR) assisted driving. In AR assisted driving, physical environment scenes are first captured by Internet of Vehicles (IoVs) and are uploaded to the Metaverse. A central Metaverse Map Service Provider (MMSP) will aggregate information from all IoVs to develop a central Metaverse Map. Information from the Metaverse Map can then be downloaded into individual IoVs on demand and be delivered as AR scenes to the driver. However, the growing interest in developing AR assisted driving applications which relies on digital twinning invites adversaries. These adversaries may place physical adversarial patches on physical world objects such as cars, signboards, or on roads, seeking to contort the virtual world digital twin. Hence, there is a need to detect these physical world adversarial patches. Nevertheless, as real-time, accurate detection of adversarial patches is compute-intensive, these physical world scenes have to be offloaded to the Metaverse Map Base Stations (MMBS) for computation. Hence in our work, we considered an environment with moving Internet of Vehicles (IoV), uploading real-time physical world scenes to the MMBSs. We formulated a realistic joint variable optimization problem where the MMSPs' objective is to maximize adversarial patch detection mean average precision (mAP), while minimizing the computed AR scene up-link transmission latency and IoVs' up-link transmission idle count, through optimizing the IoV-MMBS allocation and IoV up-link scene resolution selection. We proposed a Heterogeneous Action Proximal Policy Optimization (HAPPO) (discrete-continuous) algorithm to tackle the proposed problem. Extensive experiments shows HAPPO outperforms baseline models when compared against key metrics.

We study reward-free reinforcement learning (RL) with linear function approximation, where the agent works in two phases: (1) in the exploration phase, the agent interacts with the environment but cannot access the reward; and (2) in the planning phase, the agent is given a reward function and is expected to find a near-optimal policy based on samples collected in the exploration phase. The sample complexities of existing reward-free algorithms have a polynomial dependence on the planning horizon, which makes them intractable for long planning horizon RL problems. In this paper, we propose a new reward-free algorithm for learning linear mixture Markov decision processes (MDPs), where the transition probability can be parameterized as a linear combination of known feature mappings. At the core of our algorithm is uncertainty-weighted value-targeted regression with exploration-driven pseudo-reward and a high-order moment estimator for the aleatoric and epistemic uncertainties. When the total reward is bounded by $1$, we show that our algorithm only needs to explore $\tilde O( d^2\varepsilon^{-2})$ episodes to find an $\varepsilon$-optimal policy, where $d$ is the dimension of the feature mapping. The sample complexity of our algorithm only has a polylogarithmic dependence on the planning horizon and therefore is ``horizon-free''. In addition, we provide an $\Omega(d^2\varepsilon^{-2})$ sample complexity lower bound, which matches the sample complexity of our algorithm up to logarithmic factors, suggesting that our algorithm is optimal.

Conversational interfaces provide a flexible and easy way for users to seek information that may otherwise be difficult or inconvenient to obtain. However, existing interfaces generally fall into one of two categories: FAQs, where users must have a concrete question in order to retrieve a general answer, or dialogs, where users must follow a predefined path but may receive a personalized answer. In this paper, we introduce Conversational Tree Search (CTS) as a new task that bridges the gap between FAQ-style information retrieval and task-oriented dialog, allowing domain-experts to define dialog trees which can then be converted to an efficient dialog policy that learns only to ask the questions necessary to navigate a user to their goal. We collect a dataset for the travel reimbursement domain and demonstrate a baseline as well as a novel deep Reinforcement Figure 1: An example of the proposed task: Slice of Learning architecture for this task. Our a dialog tree (blue/gray nodes, black edges) showing results show that the new architecture combines how progressively more concrete questions could be the positive aspects of both the FAQ answered. Question a) guiding a user with a general and dialog system used in the baseline and goal through the tree, b) asking only at nodes that need achieves higher goal completion while skipping more clarification, and c) requiring no clarification and unnecessary questions.

Early stopping based on the validation set performance is a popular approach to find the right balance between under- and overfitting in the context of supervised learning. However, in reinforcement learning, even for supervised sub-problems such as world model learning, early stopping is not applicable as the dataset is continually evolving. As a solution, we propose a new general method that dynamically adjusts the update to data (UTD) ratio during training based on under- and overfitting detection on a small subset of the continuously collected experience not used for training. We apply our method to DreamerV2, a state-of-the-art model-based reinforcement learning algorithm, and evaluate it on the DeepMind Control Suite and the Atari $100$k benchmark. The results demonstrate that one can better balance under- and overestimation by adjusting the UTD ratio with our approach compared to the default setting in DreamerV2 and that it is competitive with an extensive hyperparameter search which is not feasible for many applications. Our method eliminates the need to set the UTD hyperparameter by hand and even leads to a higher robustness with regard to other learning-related hyperparameters further reducing the amount of necessary tuning.

Reaching disabilities affect the quality of life. Functional Electrical Stimulation (FES) can restore lost motor functions. Yet, there remain challenges in controlling FES to induce desired movements. Neuromechanical models are valuable tools for developing FES control methods. However, focusing on the upper extremity areas, several existing models are either overly simplified or too computationally demanding for control purposes. Besides the model-related issues, finding a general method for governing the control rules for different tasks and subjects remains an engineering challenge. Here, we present our approach toward FES-based restoration of arm movements to address those fundamental issues in controlling FES. Firstly, we present our surface-FES-oriented neuromechanical models of human arms built using well-accepted, open-source software. The models are designed to capture significant dynamics in FES controls with minimal computational cost. Our models are customisable and can be used for testing different control methods. Secondly, we present the application of reinforcement learning (RL) as a general method for governing the control rules. In combination, our customisable models and RL-based control method open the possibility of delivering customised FES controls for different subjects and settings with minimal engineering intervention. We demonstrate our approach in planar and 3D settings.

The flocking motion control is concerned with managing the possible conflicts between local and team objectives of multi-agent systems. The overall control process guides the agents while monitoring the flock-cohesiveness and localization. The underlying mechanisms may degrade due to overlooking the unmodeled uncertainties associated with the flock dynamics and formation. On another side, the efficiencies of the various control designs rely on how quickly they can adapt to different dynamic situations in real-time. An online model-free policy iteration mechanism is developed here to guide a flock of agents to follow an independent command generator over a time-varying graph topology. The strength of connectivity between any two agents or the graph edge weight is decided using a position adjacency dependent function. An online recursive least squares approach is adopted to tune the guidance strategies without knowing the dynamics of the agents or those of the command generator. It is compared with another reinforcement learning approach from the literature which is based on a value iteration technique. The simulation results of the policy iteration mechanism revealed fast learning and convergence behaviors with less computational effort.