Computation offloading and information freshness in edge-assisted networked robotic systems
Authors:
Open Access
- 4 November 2026
- journal article
- research article
- Published by Frontiers Media SA in Frontiers in Robotics and AI
- Vol. 13, 1999591
Abstract
Camera-based wireless networked control systems must move visual information across a shared wireless link before a command can be applied, so the placement of the perception computation fixes both the communication load and the freshness of the information that drives the actuator. We partition a road-following perception pipeline between a mobile robot and a remote controller at four cut points, spanning measured application payloads from a 61,570 B raw image (PUPPET) to a 4 B road-center feature (E3), and evaluate the system on a physical testbed under two controlled uplink conditions. Freshness is quantified as actuation-side Age of Information (AoI): the age of the camera observation underlying the command actually applied at the motors, reconstructed in continuous time from a single monotonic clock on the robot. Under the constrained channel, mean actuation AoI falls from 82.7 ms in PUPPET to 44.8 ms in E3, and the 95th-percentile AoI decreases from 208.7 ms to 68.7 ms. The payload–freshness relation saturates sharply. The first edge cut removes 95.6 % of the raw-image payload and captures 96.6 % – 99.6 % of the total freshness improvement, whereas a further 99.85 % reduction in payload yields only an additional 0.5 ms improvement, which remains statistically detectable after multiple-comparison correction but is between approximately 28- and 269-times smaller than the preceding PUPPET-to-E1 improvement, depending on the channel and freshness metric. A separate closed-loop campaign shows the same knee in image-plane tracking error, with the first cut capturing 96.4 % of the PUPPET-to-E3 RMSE reduction. Extending the measured payloads to a shared medium under a fixed-overhead airtime model shows that the same knee governs multi-robot scalability, but at a different cut: because a payload-independent per-frame cost dominates once an update fits in a single frame, the budget-based robot count rises 30-fold from PUPPET to E1 but only 1.8-fold from E1 to E3 against a 674-fold reduction in transmitted bytes. Actuation freshness, sustainable robot density, and raw byte volume are, therefore, exhausted in three different ways along the offloading axis, so computation placement in cooperative multi-robot deployments should be selected against the binding system constraint rather than by assuming smaller payloads are uniformly better.Keywords
This publication has 21 references indexed in Scilit:
- A computation offloading scheme based on age of information for substation power IoTAd Hoc Networks, 2025
- Task offloading and resource allocation for multi-UAV asset edge computing with multi-agent deep reinforcement learningComputing, 2025
- Age-of-Information-Based Computation Offloading and Transmission Scheduling in Mobile-Edge-Computing-Enabled IoT NetworksIEEE Internet of Things Journal, 2023
- Age of information in IoT-based networked control systems: A MAC perspectiveAutomatica, 2022
- Semantics- and Task-Oriented Scheduling for Networked Control Systems in PracticeIEEE Access, 2022
- A Critical Review of Communications in Multi-robot SystemsCurrent Robotics Reports, 2022
- Multi-Robot Coordination Analysis, Taxonomy, Challenges and Future ScopeJournal of Intelligent & Robotic Systems, 2021
- Effective age of information in real-time wireless feedback control systemsScience China Information Sciences, 2021
- Edge Computing for the Internet of Things: A Case StudyIEEE Internet of Things Journal, 2018
- Real-time status: How often should one update?2012