Autonomous safety monitoring technology capable of fall detection, mobile patrol, and rapid incident response exists commercially, but is priced between USD 11,000 and USD 75,000 per unit. This cost profile systematically excludes the schools, rural clinics, and community centers where undetected safety incidents are most consequential. A low-cost, open-source, multi-modal robotic safety platform suitable for institutional deployment in resource-constrained environments has not previously been described or evaluated in the literature. We designed, fabricated, and conducted a proof-of-concept evaluation of AURA (Affordable Unified Robotic Architecture), an open-source safety monitoring platform integrating: (i) a 16-DOF servo-driven quadruped for enclosed-space and stairwell patrol; (ii) a Pixhawk PX4-guided quadrotor for complementary aerial coverage; (iii) a YOLOv5n fall detection module running at 18 fps on a Raspberry Pi 4 edge node with no cloud dependency; and (iv) a two-channel forearm surface electromyography (sEMG) command interface supporting operators with limited motor function. All custom structural components were CNC-fabricated from 3 mm ABS sheet. Proof-of-concept evaluation was conducted in an occupied three-story secondary educational building (2,840 m²) over a six-week period under institutional ethics oversight. Fall detection was benchmarked on the publicly available UR Fall Detection Dataset. sEMG recognition was evaluated across ten healthy adult participants (n = 10; age 19–58 years). Patrol coverage and incident response were evaluated across ten patrol cycles and twenty staged fall scenarios, respectively. The fall detection module achieved F1 = 94.3% on the UR benchmark and F1 = 91.7% on a site-collected test set (n = 45 staged falls) under challenging institutional lighting, with a mean end-to-end alert latency of 1.8 ± 0.3 s. The sEMG interface achieved mean five-class gesture recognition accuracy of 94.2 ± 2.7% across participants; two participants with self-reported reduced grip strength achieved 93.1–94.4% accuracy with the EMG interface versus 86.7–88.2% with keyboard input. Combined ground-aerial patrol coverage reached 93.6 ± 2.1% versus 81.4 ± 3.2% ground-only and 64.7 ± 4.8% aerial-only. Mean end-to-end incident response time was 24.1 ± 6.4 s with the drone in flight and 31.7 ± 9.2 s docked. Zero geofence violations were recorded across all aerial patrol cycles. Total dual-platform cost was USD 400, representing an 87% reduction against the nearest commercial alternative. These preliminary findings suggest that autonomous safety monitoring at proof-of-concept acceptability thresholds for fall detection, patrol coverage, and incident response may be achievable at a cost accessible to under-resourced institutions, warranting further investigation at larger scale. These proof-of-concept findings support the feasibility of the proposed architecture; multi-site longitudinal validation with larger participant samples is required before clinical deployment. All hardware designs, firmware configurations, trained model weights, and fabrication documentation are released under open-source licenses.