Sceniris: A Fast Procedural Scene Generation Framework

TL;DR

Sceniris introduces a GPU-accelerated, batched procedural scene generation framework that substantially increases throughput for generating collision-free, robot-reachability-plausible 3D scenes. It extends Scene Synthesizer with batched scene representations, caching, GPU-based collision checking via cuRobo, and expanded spatial relationships, including reachability checks via RM4D. The approach achieves significant speedups (e.g., 311x cold-start, up to ~2936x with warm-start) and scales to hundreds of thousands of scenes per batch, enabling rapid data generation for embodied AI and 3D perception tasks. The work provides practical improvements and a public implementation to support high-throughput synthetic scene generation at scale.

Abstract

Synthetic 3D scenes are essential for developing Physical AI and generative models. Existing procedural generation methods often have low output throughput, creating a significant bottleneck in scaling up dataset creation. In this work, we introduce Sceniris, a highly efficient procedural scene generation framework for rapidly generating large-scale, collision-free scene variations. Sceniris also provides an optional robot reachability check, providing manipulation-feasible scenes for robot tasks. Sceniris is designed for maximum efficiency by addressing the primary performance limitations of the prior method, Scene Synthesizer. Leveraging batch sampling and faster collision checking in cuRobo, Sceniris achieves at least 234x speed-up over Scene Synthesizer. Sceniris also expands the object-wise spatial relationships available in prior work to support diverse scene requirements. Our code is available at https://github.com/rai-inst/sceniris

Figures (7)

• Figure 1: Sceniris is a procedural scene generation tool that provides fast, scalable, and physical-AI plausible scene generation. We leverage GPU to accelerate the collision checking and robot reachability check.
• Figure 2: Sceniris takes a configuration and generates a batch of scenes efficiently. Alternatively, Sceniris can generate a scene by adding objects step-by-step, but the efficiency is not optimized.
• Figure 3: Examples of additional spatial relationships supported by Sceniris
• Figure 4: (a) The benchmark scene, simplified for visualization. (b) Execution time. (c) The number of valid environments generated per second. Sceniris handles 311 and 6,321 times more environments per second than 10 multi-processed Scene Synthesizer, under cold start and warm start, respectively.
• Figure 5: Total time spent on key steps in Scene Synthesizer (1 scene) and Sceniris (ours, 1024 scenes). Both approaches use the same scene configuration in Section \ref{['sec:batch_scene_generation']}. Regions in green show the time spent on placing objects.