Offline Zero-Shot Reinforcement Learning requires an agent to solve unseen tasks using only a fixed offline dataset without explicit rewards. A central challenge is learning representations that capture both high-level long-term planning and low-level physical dynamics. We propose a novel framework, Dynamics-Aware Planning Representation (DAPR), which disentangles these two aspects via complementary contrastive objectives. Specifically, DAPR learns goal-oriented planning directions and local dynamics-consistent directions in the latent space. By jointly enforcing these constraints, DAPR yields representations that balance “where to go” with “how to move.” Experiments on standard locomotion benchmarks (Walker, Cheetah, Quadruped) demonstrate that DAPR consistently improves performance and generalization over strong baselines, achieving substantial gains on precision demanding tasks.

Everything can be mobile, from end-hosts to applications. The need for mobility management is rising rapidly due to the increasing number of mobile devices and the advent of new types of mobile object, such as Internet of Things devices, connected vehicles, wearable devices, and virtual machines. In order to support various types of mobility, we propose an integrated and distributed mobility management approach, called Mobility of Everything (MoE). The MoE approach uses an object ID whose value is used to determine a specific edge switch that an object is connected with. In addition, the approach distributes binding information of an object ID and an IP address of edge switch throughout the network. We implement our approach on both an emulation environment and a testbed with software switches. Our evaluation results demonstrate that the MoE approach achieves seamless and scalable handover of end-hosts and applications.