Research Short Robotics (cs.RO) arXiv:arXiv:2607.01088v1

ROSA: A Robotics Foundation Model Serving System for Robot Factories

Wenqi Jiang, Jason Clemons, Rowland O'Flaherty, Hugo Hadfield, Alperen Degirmenci, Shuran Song, Yashraj Narang, Christos Kozyrakis: ROSA: A Robotics Foundation Model Serving System for Robot Factories. CoRR abs/arXiv:2607.01088

##robotics ##robotics-foundation-models

↑ Uploaded by SheQAI Grounding report · 7 claims traced arXiv:arXiv:2607.01088v1 doi:https://doi.org/10.48550/arXiv.2607.01088 ▶ 8

⧉ Cite

APA

Jiang, W., Clemons, J., O'Flaherty, R., Hadfield, H., Degirmenci, A., Song, S., Narang, Y., & Kozyrakis, C. (n.d.). ROSA: A Robotics Foundation Model Serving System for Robot Factories. arXiv:arXiv:2607.01088. https://doi.org/https://doi.org/10.48550/arXiv.2607.01088

BibTeX

@article{jiang,
  title  = {ROSA: A Robotics Foundation Model Serving System for Robot Factories},
  author = {Wenqi Jiang and Jason Clemons and Rowland O'Flaherty and Hugo Hadfield and Alperen Degirmenci and Shuran Song and Yashraj Narang and Christos Kozyrakis},
  eprint = {arXiv:2607.01088},
  archivePrefix = {arXiv},
  primaryClass = {cs.RO},
  doi    = {https://doi.org/10.48550/arXiv.2607.01088},
}

📄 Transcript

RFM serving assumes one robot, one model.

Robotics foundation models are making general-purpose factory robots

practical, but the serving systems behind them rest

on a single-robot, single-model assumption.

The key idea is ROSA, a shared GPU-pool

serving architecture where a fleet of robots reaches

server-class GPUs over the network.

On synthetic large-scale workloads, ROSA improves factory productivity

by up to twelve point zero six times

over dedicated serving, and up to two point

four four times over shared-server baselines.

Read the paper for details.

✓ Claims & sources (7)

Each claim in this video, with the span of the paper it comes from.

Key point RFM serving treats inference as edge computing on an on-robot or dedicated nearby GPU.

inference is treated as an edge-computing problem handled by an on-robot or dedicated nearby GPU, and the serving objective is to minimizing the latency of a single action model

Key point Reliable factory execution needs multi-model pipelines, not one low-latency action model.

reliable execution requires multi-model pipelines (planning, safety, task-monitoring) and that factories care about aggregate productivity, not per-request latency

Key point ROSA serves a shared server-class GPU pool for a fleet of robots over the network.

ROSA adopts shared GPU-pool serving, allowing a fleet of robots to access powerful server-class GPUs over the network in order to improve inference performance, battery duration, and GPU utilization

Key point Scheduling maximizes SLO-qualified factory productivity, not per-request latency.

ROSA uses factory-objective-driven scheduling to maximize SLO-qualified factory productivity rather than minimizing individual request latency

Key point Improves factory productivity by up to 12.06 times over dedicated serving systems.

The results show that ROSA improves factory productivity by up to 12.06 over conventional dedicated serving systems

Key point A declarative interface specifies model composition, invocation ratios, and P99 latency SLOs.

The declarative interface specifies per-task model composition, component dependencies and invocation ratios

Key point Also gains up to 2.44 times over shared-server baselines on the same workloads.

ROSA improves factory productivity by up to 12.06x over the best dedicated serving baseline and up to 2.44x over shared-server baselines

Comments (0)

No comments yet.

ROSA: A Robotics Foundation Model Serving System for Robot Factories

Related videos