Driving · 2026-10-14

RAN-JEPA: An Action-Conditioned World Model for Energy-efficient Mobile Network Control

Binghan Wu, Kai Zeng, Kai Yang, Ya-Qin Zhang, Joseph Sifakis, Ye Ouyang

AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.

TL;DR

RAN-JEPA is an action-conditioned world model for energy-efficient mobile network (RAN) control that addresses an observation–intervention asymmetry. It first learns a predictive latent representation from observation-only telemetry (and adapts it to a target deployment), freezes the encoder, then learns action-conditioned latent dynamics from controlled sleep/wake trajectories. Experiments use public operator data plus a real 4-cell 5G testbed.

Why it matters

Operational mobile networks provide abundant passive telemetry but relatively scarce controlled action–response data. The authors propose an observation-first, action-later training framework that can support what-if evaluation and closed-loop energy-aware control without requiring the model to learn representations and intervention effects from the same limited dataset.

Method

Pretrain on observation-only PM sequences using a JEPA-style predictive latent objective; adapt the encoder to a target domain with observation-only data; keep only the context encoder at the end of stage 1.
Freeze the encoder and post-train a new action-conditioned latent rollout predictor plus downstream heads from controlled sleep/wake trajectories; the rollout uses planned action plans while future environment variables are not provided as oracle inputs.
Predict five KPI heads per cell (UE count, DL/UL PRB utilization, DL/UL data volume) and a direction head; evaluate by what-if querying under different candidate action plans and by using predictions inside a closed-loop sleep controller.

Limitation

Experiments are limited to a four-cell controlled testbed and use sleep ratio as an operational proxy for energy-saving aggressiveness; they should not be interpreted as network-wide energy savings or as performance guarantees for all commercial deployments.

Abstract (from arXiv)

Operational mobile networks are rich in observations but poor in interventions: they continuously generate large volumes of passive telemetry, while controlled action--response trajectories remain comparatively scarce and costly to collect. This observation--intervention asymmetry is a central obstacle to building practical world models for RAN control. We address this problem with RAN-JEPA, an action-conditioned world model that separates representation learning from intervention learning. The model first learns a predictive latent representation from observation-only network data, adapts that representation to the target deployment domain, and then learns action-conditioned latent dynamics from controlled sleep/wake trajectories while keeping the encoder fixed. We evaluate RAN-JEPA using public operator performance data for pretraining and a real-equipment four-cell 5G testbed for domain adaptation, post-training, and control. Across four future 15-minute intervals, RAN-JEPA achieves lower action-conditioned prediction error than a GPT-2 world-model baseline and a mechanistic rule, while remaining close to a strong persistence reference. What-if evaluation further shows that the model produces distinct and coherent future trajectories under matched SLEEP and KEEP plans. In closed-loop control, RAN-JEPA consistently supports higher sleep ratios than the two baselines at comparable congestion cost. These results show that observation-first, action-conditioned learning can turn abundant passive telemetry and limited intervention data into a practical predictive model for energy-aware RAN control.

Publication record

Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.

First public release: 2026-10-13 · Indexed by Pub-AI: 2026-10-14 UTC

preprint