Robot learning adapts updates to unstable mobile networks
Network-Co-Designed Dynamic Sparsification for Dependable Distributed Robot Learning over 3GPP Networks
Networking and Internet Architecture
Summary
When robot teams learn together over mobile networks, their communication can be slowed or lost due to network problems like low bandwidth or delays. The authors created a system that shrinks the amount of data robots send based on current network conditions and time limits, helping training continue smoothly. They tested two ways to link the learning process with network controls and showed it works well even when network quality changes a lot. This means robots can learn reliably without sending too much data over shaky connections.
What this means in practice
- •For robotics system integrators: Reduce communication demands for multi-robot fleets training together over variable cellular networks to maintain efficient learning progress.
- •For mobile network engineers: Incorporate AI-specific feedback loops in 3GPP-based networks to better manage bandwidth for distributed machine learning applications in robotics.
Authors
Géza Szabó, Boldizsár Bunda
Abstract
Robot fleets performing distributed learning often face volatile uplink conditions like bandwidth variation, packet loss, and latency spikes that delay or destabilize training. We propose a network-co-designed dynamic sparsification loop that couples model-update compression to network-facing byte budgets and time constraints. We describe two realizations: (i) a standards-aligned SA6/5GC mapping that leverages CAPIF with NEF/PCF/NWDAF, enriched with AI-specific semantics; and (ii) a VASEAL-style role-reversed loop where the learning system exposes learning-state feedback and receives bounded, time-limited control intents. We implement the mechanism in an asynchronous federated reinforcement learning framework with heterogeneous simulated quadruped robots and reproducible uplink impairments via emulated network links. Results indicate that substantial communication reduction is achievable under dynamic constraints without materially degrading learning progress in the studied setting.