Research map / Reinforcement learning / Visual and representation RL
Boba: Batched Simulation for Physics-Based Gaussian Digital Twins
Yihan Pang, Hanxiao Jiang, Sushant Kondguli, Sarita Adve, Shenlong Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| apply Boba's local optimization stack (coarsening, skinning, rendering) to PhysTwin | PhysTwin · Single-instance latency (Orin) | latency (ms): 330.9 -> 32.5 | Improved |
| offload simulation and skinning to a server (Boba-Distributed) | Boba-Local · Single-instance latency (Orin) | latency (ms): 32.5 -> 25.1 | Improved |
| offload simulation and skinning to a server (Boba-Distributed) | Boba-Local · Incremental power (Orin) | power (mW): 6112 -> 4758 | Improved |
| add physical simulation optimization (coarsening, pruning, locality) (ablation) | PhysTwin · Single-instance throughput (Orin) | FPS: 3.0 -> 8.1 | Improved |
| add skinning optimization on top of simulation optimization (ablation) | Boba-Local · Single-instance throughput (Orin) | FPS: 8.1 -> 21.5 | Improved |
| add rendering/visualization optimization (ablation) | Boba-Local · Single-instance throughput (Orin) | FPS: 21.5 -> 30.8 | Improved |
| add template sharing in batched execution (ablation) | PhysTwin-MultiProc · Batch capacity (average) | instances: 11.0 -> 369.2 | Improved |
| add template sharing in batched execution (ablation) | PhysTwin-MultiProc · Batched throughput (RTX 4090) | FPS: 44.7 -> 2392.8 | Improved |
| add atomic-free force accumulation (ablation) | Boba-Batched (template-share only) · Batched throughput (RTX 4090) | FPS: 2392.8 -> 3309.9 | Improved |
| add atomic-free force accumulation (ablation) | Boba-Batched (template-share only) · Batch capacity (average) | instances: 369.2 -> 368.3 | No clear effect |
| use Boba-Local optimizations instead of PhysTwin | PhysTwin · Reconstruction & resimulation | PSNR: 28.7 -> 28.4 | No clear effect |
| use Boba-Local optimizations instead of PhysTwin | PhysTwin · Reconstruction & resimulation | LPIPS: 0.025 -> 0.031 | Hurt |
Closest papers by meaning
- PhySPRING: Structure-Preserving Reduction of Physics-Informed Digital Twins via Graph Neural Networks (NeurIPS 2026 · similarity 0.89)
- RoboTwin: Dual-Arm Robot Benchmark with Generative Digital Twins (CVPR 2025 · similarity 0.85)
- Simulation-Ready Compositional 3D Scene Reconstruction from a Single Image (NeurIPS 2026 · similarity 0.85)
- RoboScape: Physics-informed Embodied World Model (NeurIPS 2025 · similarity 0.85)
- URDF-Anything+: End-to-End Generation for Simulation-Ready Articulated Assets (NeurIPS 2026 · similarity 0.84)
- PhysGen3D: Crafting a Miniature Interactive World from a Single Image (CVPR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
