Mixture-of-World Models: Scaling Multi-Task Reinforcement Learning with Modular Latent Dynamics

A fundamental challenge in multi-task reinforcement learning (MTRL) is achieving sample efficiency in visual domains where tasks exhibit substantial heterogeneity in both observations and dynamics. Model-based reinforcement learning offers a promising path to improved sample efficiency through world models, but standard monolithic architectures struggle to capture diverse task dynamics, resulting in poor reconstruction and prediction accuracy. We introduce Mixture-of-World Models (MoW), a scalable architecture that combines modular variational autoencoders for task-adaptive visual compression, a hybrid Transformer-based dynamics model with task-conditioned experts and a shared backbone, and a gradient-based task clustering strategy for efficient parameter allocation. On the Atari 100k benchmark, a single MoW agent trained once on 26 Atari games achieves a mean human-normalized score of 110.4%, competitive with the score of 114.2% achieved by STORM, an ensemble of 26 task-specific models, while using 50% fewer parameters. On Meta-World, MoW achieves a 74.5% average success rate within 300 thousand environment steps, establishing a new state of the art. These results demonstrate that MoW provides a scalable and parameter-efficient foundation for generalist world models.

Multi-agent Embodied AI: Advances and Future Directions

Embodied artificial intelligence (Embodied AI) plays a pivotal role in the application of advanced technologies in the intelligent era, where AI systems are integrated with physical bodies that enable them to perceive, reason, and interact with their environments. Through the use of sensors for input and actuators for action, these systems can learn and adapt based on real-world feedback, allowing them to perform tasks effectively in dynamic and unpredictable environments. As techniques such as deep learning (DL), reinforcement learning (RL), and large language models (LLMs) mature, embodied AI has become a leading field in both academia and industry, with applications spanning robotics, healthcare, transportation, and manufacturing. However, most research has focused on single-agent systems that often assume static, closed environments, whereas real-world embodied AI must navigate far more complex scenarios. In such settings, agents must not only interact with their surroundings but also collaborate with other agents, necessitating sophisticated mechanisms for adaptation, real-time learning, and collaborative problem-solving. Despite increasing interest in multi-agent systems, existing research remains narrow in scope, often relying on simplified models that fail to capture the full complexity of dynamic, open environments for multi-agent embodied AI. Moreover, no comprehensive survey has systematically reviewed the advancements in this area. As embodied AI rapidly evolves, it is crucial to deepen our understanding of multi-agent embodied AI to address the challenges presented by real-world applications. To fill this gap and foster further development in the field, this paper reviews the current state of research, analyzes key contributions, and identifies challenges and future directions, providing insights to guide innovation and progress in this field.

Time-optimal Flight in Cluttered Environments via Safe Reinforcement Learning

This paper addresses the problem of guiding a quadrotor through a predefined sequence of waypoints in cluttered environments, aiming to minimize the flight time while avoiding collisions. Previous approaches either suffer from prolonged computational time caused by solving complex non-convex optimization problems or are limited by the inherent smoothness of polynomial trajectory representations, thereby restricting the flexibility of movement. In this work, we present a safe reinforcement learning approach for autonomous drone racing with time-optimal flight in cluttered environments. The reinforcement learning policy, trained using safety and terminal rewards specifically designed to enforce near time-optimal and collision-free flight, outperforms current state-of-the-art algorithms. Additionally, experimental results demonstrate the efficacy of the proposed approach in achieving both minimum flight time and obstacle avoidance objectives in complex environments, with a commendable $66.7\%$ success rate in unseen, challenging settings.

Learning Hybrid Policies for MPC with Application to Drone Flight in Unknown Dynamic Environments

In recent years, drones have found increased applications in a wide array of real-world tasks. Model predictive control (MPC) has emerged as a practical method for drone flight control, owing to its robustness against modeling errors/uncertainties and external disturbances. However, MPC's sensitivity to manually tuned parameters can lead to rapid performance degradation when faced with unknown environmental dynamics. This paper addresses the challenge of controlling a drone as it traverses a swinging gate characterized by unknown dynamics. This paper introduces a parameterized MPC approach named hyMPC that leverages high-level decision variables to adapt to uncertain environmental conditions. To derive these decision variables, a novel policy search framework aimed at training a high-level Gaussian policy is presented. Subsequently, we harness the power of neural network policies, trained on data gathered through the repeated execution of the Gaussian policy, to provide real-time decision variables. The effectiveness of hyMPC is validated through numerical simulations, achieving a 100\% success rate in 20 drone flight tests traversing a swinging gate, demonstrating its capability to achieve safe and precise flight with limited prior knowledge of environmental dynamics.