Replicate What Varies: Resource Sharing for Domain-Randomized Embodied Simulation

Patchwork is a resource‑sharing system for domain‑randomized embodied simulation that separates a shared template of common resources from patches that record environment‑specific mutations. By sharing compiled model structure and immutable assets while keeping mutable parameters private, Patchwork reduces CPU memory usage dramatically—up to 92% at 32 replicas—compared to running each environment in its own process. The approach was implemented on robosuite and MuJoCo and evaluated on LIBERO‑plus and RoboCasa365 workloads, demonstrating that memory growth shifts toward template diversity rather than instance count.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.

• agents
• robotics