Foundation Models for Robotics Show Strengths and Gaps
A benchmark of VLA models across 18 tasks reveals strong object generalization but weak force control and long-horizon planning.
Underlying Paper
Foundation Models for Robotic Manipulation: A Survey and Benchmark
We present a comprehensive evaluation of vision-language-action (VLA) foundation models for robotic manipulation across 18 tasks spanning dexterous grasping, tool use, and multi-step assembly. Our benchmark introduces standardized evaluation protocols and reveals that while VLA models show strong zero-shot generalization to novel objects, they consistently struggle with precise force control and long-horizon planning.
The pursuit of scalable generative models has driven a wave of architectural innovation, yet the quadratic cost of attention in transformer-based diffusion models remains a fundamental bottleneck. This paper introduces a compelling alternative: replacing the attention backbone entirely with structured state space models (SSMs).
Core Contribution
The authors demonstrate that Mamba-style SSMs can serve as drop-in replacements for attention layers in the U-Net architecture commonly used for diffusion. The key insight is that the selective scan mechanism of modern SSMs naturally captures the multi-scale spatial dependencies required for high-quality image generation.
Technical Approach
The architecture, dubbed DiS (Diffusion with State Spaces), modifies the standard DiT (Diffusion Transformer) by replacing each attention block with a bidirectional SSM layer. The authors introduce a novel "cross-scan" strategy that processes image patches along four spatial directions simultaneously, aggregating the results to capture both local texture and global structure.
Results and Analysis
On ImageNet 256×256 unconditional generation, DiS achieves an FID of 2.67, comparable to DiT-XL/2 (FID 2.27) while requiring 3.2× fewer FLOPs per denoising step. The gap narrows further at 512×512 resolution, where DiS achieves FID 3.41 vs. DiT's 3.04 — a marginal quality difference that may be acceptable given the substantial computational savings.
Training convergence is notably faster: DiS reaches its best FID in approximately 400K steps compared to DiT's 700K steps under identical training budgets. The linear-time scaling also enables generation at resolutions the transformer variant cannot practically reach without additional engineering.
Evidence Box
moderateKey Claims
- •VLA models generalize to novel objects in zero-shot
- •Consistent failures in precise force control
- •Long-horizon planning remains challenging
Key Results
- •78% success on single-step novel object grasping
- •23% success on multi-step assembly tasks
- •41% success on tasks requiring force modulation
Limitations & Caveats
- •Benchmark limited to tabletop manipulation
- •Real-world evaluation on single robot platform
- •No evaluation of sim-to-real transfer
