Decision-making for urban autonomous driving is challenging due to the stochastic nature of interactive traffic participants and the complexity of road structures. Although reinforcement learning (RL)-based decision-making scheme is promising to handle urban driving scenarios, it suffers from low sample efficiency and poor adaptability. In this paper, we propose Scene-Rep Transformer to improve the RL decision-making capabilities with better scene representation encoding and sequential predictive latent distillation. Specifically, a multi-stage Transformer (MST) encoder is constructed to model not only the interaction awareness between the ego vehicle and its neighbors but also intention awareness between the agents and their candidate routes. A sequential latent Transformer (SLT) with self-supervised learning objectives is employed to distill the future predictive information into the latent scene representation, in order to reduce the exploration space and speed up training. The final decision-making module based on soft actor-critic (SAC) takes as input the refined latent scene representation from the Scene-Rep Transformer and outputs driving actions. The framework is validated in five challenging simulated urban scenarios with dense traffic, and its performance is manifested quantitatively by the substantial improvements in data efficiency and performance in terms of success rate, safety, and efficiency. The qualitative results reveal that our framework is able to extract the intentions of neighbor agents to help make decisions and deliver more diversified driving behaviors.

In this paper we treat optimal trajectory planning for an autonomous vehicle (AV) operating in dense traffic, where vehicles closely interact with each other. To tackle this problem, we present a novel framework that couples trajectory prediction and planning in multi-agent environments, using distributed model predictive control. A demonstration of our framework is presented in simulation, employing a trajectory planner using non-linear model predictive control. We analyze performance and convergence of our framework, subject to different prediction errors. The results indicate that the obtained locally optimal solutions are improved, compared with decoupled prediction and planning.

Large-scale models require substantial computational resources for analysis and studying treatment conditions. Specifically, estimating treatment effects using simulations may require a lot of infeasible resources to allocate at every treatment condition. Therefore, it is essential to develop efficient methods to allocate computational resources for estimating treatment effects. Agent-based simulation allows us to generate highly realistic simulation samples. FRED (A Framework for Reconstructing Epidemiological Dynamics) is an agent-based modeling system with a geospatial perspective using a synthetic population constructed based on the U.S. census data. Given its synthetic population, FRED simulations present a baseline for comparable results from different treatment conditions and treatment conditions. In this paper, we show three other methods for estimating treatment effects. In the first method, we resort to brute-force allocation, where all treatment conditions have an equal number of samples with a relatively large number of simulation runs. In the second method, we try to reduce the number of simulation runs by customizing individual samples required for each treatment effect based on the width of confidence intervals around the mean estimates. In the third method, we use a regression model, which allows us to learn across the treatment conditions such that simulation samples allocated for a treatment condition will help better estimate treatment effects in other conditions. We show that the regression-based methods result in a comparable estimate of treatment effects with less computational resources. The reduced variability and faster convergence of model-based estimates come at the cost of increased bias, and the bias-variance trade-off can be controlled by adjusting the number of model parameters (e.g., including higher-order interaction terms in the regression model).

The increasing trend in adopting electric vehicles (EVs) will significantly impact the residential electricity demand, which results in an increased risk of transformer overload in the distribution grid. To mitigate such risks, there are urgent needs to develop effective EV charging controllers. Currently, the majority of the EV charge controllers are based on a centralized approach for managing individual EVs or a group of EVs. In this paper, we introduce a decentralized Multi-agent Reinforcement Learning (MARL) charging framework that prioritizes the preservation of privacy for EV owners. We employ the Centralized Training Decentralized Execution-Deep Deterministic Policy Gradient (CTDE-DDPG) scheme, which provides valuable information to users during training while maintaining privacy during execution. Our results demonstrate that the CTDE framework improves the performance of the charging network by reducing the network costs. Moreover, we show that the Peak-to-Average Ratio (PAR) of the total demand is reduced, which, in turn, reduces the risk of transformer overload during the peak hours.

Large-scale simulation with realistic nonlinear dynamic models is crucial for algorithms development for swarm robotics. However, existing platforms are mainly developed based on Object-Oriented Programming (OOP) and either use simple kinematic models to pursue a large number of simulating nodes or implement realistic dynamic models with limited simulating nodes. In this paper, we develop a simulator based on Data-Oriented Programming (DOP) that utilizes GPU parallel computing to achieve large-scale swarm robotic simulations. Specifically, we use a multi-process approach to simulate heterogeneous agents and leverage PyTorch with GPU to simulate homogeneous agents with a large number. We test our approach using a nonlinear quadrotor model and demonstrate that this DOP approach can maintain almost the same computational speed when quadrotors are less than 5,000. We also provide two examples to present the functionality of the platform.

We propose a novel master-slave architecture to solve the top-$K$ combinatorial multi-armed bandits problem with non-linear bandit feedback and diversity constraints, which, to the best of our knowledge, is the first combinatorial bandits setting considering diversity constraints under bandit feedback. Specifically, to efficiently explore the combinatorial and constrained action space, we introduce six slave models with distinguished merits to generate diversified samples well balancing rewards and constraints as well as efficiency. Moreover, we propose teacher learning based optimization and the policy co-training technique to boost the performance of the multiple slave models. The master model then collects the elite samples provided by the slave models and selects the best sample estimated by a neural contextual UCB-based network to make a decision with a trade-off between exploration and exploitation. Thanks to the elaborate design of slave models, the co-training mechanism among slave models, and the novel interactions between the master and slave models, our approach significantly surpasses existing state-of-the-art algorithms in both synthetic and real datasets for recommendation tasks. The code is available at: \url{https://github.com/huanghanchi/Master-slave-Algorithm-for-Top-K-Bandits}.

The comprehension of how local interactions arise in global collective behavior is of utmost importance in both biological and physical research. Traditional agent-based models often rely on static rules that fail to capture the dynamic strategies of the biological world. Reinforcement learning has been proposed as a solution, but most previous methods adopt handcrafted reward functions that implicitly or explicitly encourage the emergence of swarming behaviors. In this study, we propose a minimal predator-prey coevolution framework based on mixed cooperative-competitive multiagent reinforcement learning, and adopt a reward function that is solely based on the fundamental survival pressure, that is, prey receive a reward of $-1$ if caught by predators while predators receive a reward of $+1$. Surprisingly, our analysis of this approach reveals an unexpectedly rich diversity of emergent behaviors for both prey and predators, including flocking and swirling behaviors for prey, as well as dispersion tactics, confusion, and marginal predation phenomena for predators. Overall, our study provides novel insights into the collective behavior of organisms and highlights the potential applications in swarm robotics.

Environment prediction frameworks are integral for autonomous vehicles, enabling safe navigation in dynamic environments. LiDAR generated occupancy grid maps (L-OGMs) offer a robust bird's eye-view scene representation that facilitates joint scene predictions without relying on manual labeling unlike commonly used trajectory prediction frameworks. Prior approaches have optimized deterministic L-OGM prediction architectures directly in grid cell space. While these methods have achieved some degree of success in prediction, they occasionally grapple with unrealistic and incorrect predictions. We claim that the quality and realism of the forecasted occupancy grids can be enhanced with the use of generative models. We propose a framework that decouples occupancy prediction into: representation learning and stochastic prediction within the learned latent space. Our approach allows for conditioning the model on other available sensor modalities such as RGB-cameras and high definition maps. We demonstrate that our approach achieves state-of-the-art performance and is readily transferable between different robotic platforms on the real-world NuScenes, Waymo Open, and a custom dataset we collected on an experimental vehicle platform.

A reliable communication network is essential for multiple UAVs operating within obstacle-cluttered environments, where limited communication due to obstructions often occurs. A common solution is to deploy intermediate UAVs to relay information via a multi-hop network, which introduces two challenges: (i) how to design the structure of multi-hop networks; and (ii) how to maintain connectivity during collaborative motion. To this end, this work first proposes an efficient constrained search method based on the minimum-edge RRT$^\star$ algorithm, to find a spanning-tree topology that requires a less number of UAVs for the deployment task. To achieve this deployment, a distributed model predictive control strategy is proposed for the online motion coordination. It explicitly incorporates not only the inter-UAV and UAV-obstacle distance constraints, but also the line-of-sight (LOS) connectivity constraint. These constraints are well-known to be nonlinear and often tackled by various approximations. In contrast, this work provides a theoretical guarantee that all agent trajectories are ensured to be collision-free with a team-wise LOS connectivity at all time. Numerous simulations are performed in 3D valley-like environments, while hardware experiments validate its dynamic adaptation when the deployment position changes online.

End-to-end learning has emerged as a major paradigm for developing autonomous systems. Unfortunately, with its performance and convenience comes an even greater challenge of safety assurance. A key factor of this challenge is the absence of the notion of a low-dimensional and interpretable dynamical state, around which traditional assurance methods revolve. Focusing on the online safety prediction problem, this paper proposes a configurable family of learning pipelines based on generative world models, which do not require low-dimensional states. To implement these pipelines, we overcome the challenges of learning safety-informed latent representations and missing safety labels under prediction-induced distribution shift. These pipelines come with statistical calibration guarantees on their safety chance predictions based on conformal prediction. We perform an extensive evaluation of the proposed learning pipelines on two case studies of image-controlled systems: a racing car and a cartpole.