Autoregressive Policies Achieve Real-Time Execution in Vision-Language-Action Models
Researchers demonstrate that autoregressive policies can execute in real-time for robotic control by adjusting tokenization and applying constrained decoding, addressing a gap where prior work focused mainly on diffusion policies. Autoregressive policies are typically slower than diffusion approaches in synchronous inference, making real-time capability particularly valuable. The findings suggest autoregressive policies remain competitive for practical robot deployment while offering advantages like faster training convergence and better instruction-following generalization.
A new study on arXiv shows that autoregressive policies—machine learning models that generate action sequences token-by-token—can achieve real-time execution speeds necessary for practical robotic systems. The researchers address a technical challenge: while diffusion-based policies have dominated recent real-time execution research, autoregressive policies have been overlooked despite their slower rollout speed. By modifying the tokenization horizon and implementing constrained decoding techniques, the authors guarantee strict latency bounds and enable multi-trajectory decoding to optimize task performance. Experiments across both simulated and real-world robotic environments show that autoregressive policies consistently outperform equivalent-level flow-matching policies while achieving significantly faster task completion from synchronous inference. The work highlights that autoregressive approaches retain inherent advantages including faster convergence during training and superior generalization to new instructions, positioning them as a viable and competitive option for real-time robotic control.
What's missing
The study does not specify which real-world robotic tasks were tested, the magnitude of performance improvements over baselines, or computational hardware requirements for achieving the reported latency bounds. Additionally, the paper does not discuss failure modes or scenarios where the approach may underperform.
What different sources said
- arXiv cs.AICenter
Real-Time Execution with Autoregressive Policies
Related
Topology-Aware Thermodynamics Improves DNA Probe Specificity Design
Researchers developed a new framework for designing DNA probes that accounts for the spatial organization of matched sequences, not just overall thermodynamic stability. Traditional methods rely on scalar measures like melting temperature and free energy, which miss how mismatches are distributed along the probe. The approach could improve diagnostic accuracy in applications like HPV detection and gene expression profiling.
Study Identifies Optimal Thermal Dose for Combining Focused Ultrasound with Immunotherapy in Tumors
Researchers used multimodal PET imaging to identify an optimal thermal dose range for focused ultrasound ablation that destroys tumor tissue while preserving conditions for immunotherapy delivery. The study found that excessive heating collapses blood vessels needed for antibody access, while insufficient heating fails to adequately reduce tumor burden. The findings could guide clinical design of combination treatments pairing thermal ablation with immunotherapies.
Plant MSH1 Protein Functions as Mismatch-Directed Nuclease for Organelle Genome Maintenance
Researchers have identified the precise mechanism by which the AtMSH1 protein in Arabidopsis plants recognizes and cleaves DNA mismatches and lesions, preventing mutations in organellar genomes. The protein combines a DNA mismatch recognition module with a nuclease domain that makes staggered cuts at specific positions relative to DNA damage. This discovery explains how plants maintain unusually low mutation rates in their mitochondrial and chloroplast DNA compared to other eukaryotes.