FogROS2-Config: Optimizing Latency and Cost for Multi-Cloud Robot Applications

TL;DR

FogROS2-Config addresses the problem of per-request compute offloading for ROS2 across multiple cloud providers by modeling a latency-cost tradeoff via a relaxed optimization objective $\alpha \cdot latency(x) + (1-\alpha) \cdot cost(x)$ over hardware $x = \{CPU, memory, GPU\}$ and selecting cost-effective configurations that satisfy user constraints. It builds surrogate latency and cost functions by benchmarking a small set of edge-case servers and uses SkyPilot to deploy and manage multi-cloud instances without modifying ROS2 code. The approach is evaluated on three robotics tasks—visual SLAM, grasp planning with Dex-Net, and motion planning—showing that the optimizer's Pareto frontier closely matches ground-truth fronts and can reduce benchmarking cost by up to about 20x. The results demonstrate practical impact by enabling affordable, latency-aware cloud offloading for diverse robotic workloads.

Abstract

Cloud service providers provide over 50,000 distinct and dynamically changing set of cloud server options. To help roboticists make cost-effective decisions, we present FogROS2-Config, an open toolkit that takes ROS2 nodes as input and automatically runs relevant benchmarks to quickly return a menu of cloud compute services that tradeoff latency and cost. Because it is infeasible to try every hardware configuration, FogROS2-Config quickly samples tests a small set of edge case servers. We evaluate FogROS2-Config on three robotics application tasks: visual SLAM, grasp planning. and motion planning. FogROS2-Config can reduce the cost by up to 20x. By comparing with a Pareto frontier for cost and latency by running the application task on feasible server configurations, we evaluate cost and latency models and confirm that FogROS2-Config selects efficient hardware configurations to balance cost and latency.

Figures (6)

• Figure 1: A Sample Use Case of FogROS2-Config. With FogROS2-Config, users only need to input application-level requirements, such as latency and per-request cost. FogROS2-Config automates the cloud machine selection by modeling the latency cost tradeoff and facilities the cost-effective cloud robotics machine selection. FogROS2-Config automatically provisions the cloud machines and enables unmodified ROS2 applications to run as if all components are on the local robot.
• Figure 2: An Overview of the Benchmark and Launch Sequence of FogROS2-Config FogROS2-Config automatically computes the most cost-effective cloud machine configuration through its optimizer. FogROS2-Config's model can be reused with different cost and latency constraints without the need to rerun the benchmark. The launch sequence is elaborated in Section \ref{['sec:design:launch']} with the same sequence numbers. (1) User specifies the application level constraints; (2) FogROS2-Config finds the optimal hardware specification; (3-6) FogROS2-Config launches the cloud server, provisions the ROS2 environment, and offloads the application nodes (7) FogROS2-Config connects the robot to the cloud server.
• Figure 3: FogROS2-Config Launch Script Example. In this example, the grasp planning node is launched to the cloud with constraints for cost of less than $0.01 and 0.01 seconds of latency per request.
• Figure 4: Pareto Frontier Analysis of FogROS2-Config on Visual SLAM with Orb-SLAM and Grasp Planning with Dex-Net. We generated a ground truth Pareto frontier plot (red) along with a Pareto frontier made from the hardware configurations sampled by the optimizer (green). In both cases, we can see that both lines follow a similar staircase pattern indicating a strong fit.
• Figure 5: Cost-Latency Pareto Frontier Analysis of Motion Planning. Due to the scholastic nature of the motion planner, the apartment test scenario demonstrates a weak correlation between latency and cloud resource costs. This leads the FogROS2-Config optimizer to make sub-optimal selections.