Improving Human-swarm Interaction Through Speech Control and Peer-to-peer Micro-agent Communication (Dataset)

In this repository we present the dataset for comparison of cyb and ROS demonstrators in cluded in the conference paper "Improving Human-swarm Interaction Through Speech Control and Peer-to-peer Micro-agent Communication". A sample image of face detection with cyb is present. We set up a scenario as per following table (see CYB_ros.csv) : Metric ROS2 Solution CyB Solution (Detection time period=0.033) Notes Architecture Monolithic (1 node) Distributed (3 nodes) CyB offers greater modularity Video Resolution 640x480 (dynamic) 640x480 Both configurable Target Frame Rate 30 FPS 30 FPS Same nominal target Effective Throughput ~30 FPS ~28 FPS ROS2 slightly superior Frame Interval 30–40 ms ~33 ms CyB more precise with Detection time period=0.033 Detection Latency < 100 ms ~30–70 ms CyB faster Detection Rate ~95% of frames ~13.6% (57/420) ROS2 still superior, but gap reduced Detection Frequency Continuous per frame Every ~7–8 frames CyB more selective but consistent Memory Usage ~500MB (single process) Distributed (3 processes) CyB parallelizable Compression Native ROS2 JPEG (~12KB/frame) CyB optimized for bandwidth Communication ROS2 Topics TCP FlowProtocol ROS2 better integrated into ecosystem Startup Time ~6 seconds ~10 seconds ROS2 still faster Processing Model Synchronous Asynchronous demand-driven CyB more efficient in standby Detection Accuracy Continuous Burst clustering CyB detects in grouped sequences Resource Scaling Single-threaded Multi-process CyB scales better Detection Clustering Single detections Multiple per frame (2x duplicates) CyB generates multiple detections Performance with lpho=0.033 CyB Improvements vs First Test: Detection rate increased from 6.7% to 13.6% (+100%) from aprevious test with lpho = 0.1 More precise timing: frame interval from ~37ms to ~33ms Reduced latency: from 50–80ms to 30–70ms Faster startup: from 15s to 10s CyB Detection Pattern: Burst detection: detects faces in concentrated sequences (e.g., frames 177–216, then a pause, then 300–329) Multiple detections per frame: often two consecutive detections per face Clustered precision: once it detects, it maintains detection across several frames Comparison ROS2 Remains Superior For: Continuous detection: 95% vs 13.6% Ecosystem integration: native to robotic stack Consistent throughput: stable 30 FPS Simple deployment: single container CyB (Optimized) Offers: Optimized latency: 30–70ms vs <100ms in ROS2 Energy efficiency: selective vs continuous detection Distributed scalability: 3 nodes vs 1 monolithic Burst accuracy: when detecting, tracks over several frames Conclusion With Detection time period=0.033, CyB becomes competitive for event-driven detection rather than continuous monitoring. ROS2 is better suited for continuous surveillance. CyB (with lpho tuning) excels in reactive/event-based detection, with reduced latency and distributed scalability. Choice depends on use case: Continuous monitoring → ROS2 Reactive alerts/event-driven detection → CyB (Detection time period=0.033)