AP-Assisted VR Streaming Dataset (BRAVR)

This dataset accompanies the paper BRAVR: An AP-Assisted Online DRL Mechanism for Interactive VR Bitrate Adaptation over Wi-Fi and contains raw experimental logs collected from a real-world Wi-Fi virtual reality (VR) streaming testbed evaluating decentralized adaptive bitrate algorithms under dynamic wireless conditions and multi-user contention. The experiments were conducted using a custom fork of Air Light VR (ALVR) v20.6.0: https://github.com/miguelcUPF/BRAVR_DRL The dataset includes both single-user and multi-user evaluation scenarios. The single-user scenario consists of one VR user (user_1) moving between two locations relative to the Wi-Fi access point: Location A: near the AP (approximately −48 dBm), Location B: far from the AP (approximately −82 dBm). The trajectory comprises four consecutive 60-second stages: (i) stationary at A, (ii)moving from A to B, (iii) stationary at B, and (iv) moving from B to A. The multi-user experiments include: a near--near scenario, where both user_1 and user_2 remain close to the access point (Location A), a near--far scenario, where user_1 remains near the AP (Location A) while user_2 is located farther away under weaker channel conditions (Location B). The dataset includes experiments for multiple bitrate adaptation strategies: nest_vr/ — NeSt-VR, a heuristic adaptive bitrate algorithm for VR streaming (see NeSt-VR: An Adaptive Bitrate Algorithm for Virtual Reality Streaming over Wi-Fi) bravr_drl/ — BRAVR, the proposed access point--assisted deep reinforcement learning framework, bravr_ablated/ — an ablated BRAVR variant without access point telemetry, cbr/ — constant bitrate baselines used in the multi-user experiments: 80 Mbps target bitrate in the near--near scenario, 70 Mbps target bitrate in the near--far scenario. For all algorithms and scenarios, the dataset includes multiple evaluation sessions of approximately 240 seconds each, stored inside the corresponding evaluation/ folders. For the learning-based methods (bravr_drl/ and bravr_ablated/), the dataset additionally includes training sessions stored inside the corresponding training/ folders: in the single-user scenario, multiple sequential training sessions of approximately 240 seconds each, in the multi-user scenarios, a single longer training session of approximately 20 minutes. Learned neural network models are persisted across runs within each scenario to enable incremental online learning. Please refer to the associated paper for detailed hardware specifications, network configuration, algorithm settings, and scenario descriptions. Each run includes a raw session_log.json file generated by ALVR containing timestamped runtime events recorded during streaming sessions. These logs include: built-in ALVR streaming statistics (GraphStatistics), including end-to-end pipeline latency and its components, additional VR network statistics (GraphNetworkStatistics), including metrics such as frame span, frame inter-arrival time, video frame round-trip time (VF-RTT), packet loss, throughput, frame jitter, packet jitter, and filtered one-way delay, access point telemetry snapshots (APStats), including channel utilization, retransmissions, airtime usage, connected VR users, and modulation and coding scheme (MCS) information, environment observations (EnvironmentSnapshot) collected at each bitrate adaptation decision interval, heuristic adaptation statistics for NeSt-VR (HeuristicStats), including bitrate decisions, bitrate adjustment probabilities, estimated capacity, and latency and frame delivery statistics, reinforcement learning statistics for BRAVR (BRAVRStats), including rewards, Q-values, policy probabilities, entropy, selected actions, and temporal-difference errors. For BRAVR and its ablated variant, learned model checkpoints (.safetensors) are additionally provided. The dataset also includes the OpenWrt-based access point monitoring scripts used during experimentation. These scripts collect wireless telemetry using standard Linux/OpenWrt utilities (e.g., iw, iwinfo, ip). Telemetry is exposed through a lightweight HTTP interface and periodically retrieved by the modified ALVR implementation. The provided monitoring scripts, inside the openwrt/ folder, include: ap_monitor.sh — lightweight telemetry collection used by BRAVR, ap_monitor_bulk.sh — extended telemetry collection for monitoring and analysis purposes. Acknowledgements This work is supported by the following projects: MLDR (Chist-ERA WAI 2022) PCI2023-145958-2 (MCIU/AEI/10.13039) REALM (GA 101298050 European Union) TRUE Wi-Fi PID2024-155470NB-I00 (MICIU/AEI/10,13039/501100011033/FEDER,UE) ICREA Academia 2024 (00077 AGAUR) MdM CEX2021-001195-M (MICIU/AEI/10.13039/501100011033) Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union. Neither the European Union nor the granting authority can be held responsible for them.