What it is

Aarakshah is a semi-autonomous situational awareness and adaptive patrol system designed for persistent monitoring of large, security-sensitive environments such as urban perimeters, border regions, public venues, and critical infrastructure.

Rather than relying on static camera grids or fully manual oversight, Aarakshah combines autonomous aerial patrols with AI-driven event interpretation to provide continuous coverage, anomaly detection, and context-aware response. The system is designed to minimize human operator load while maintaining explainability, controllability, and real-time awareness.

Why I'm building it

Modern surveillance and security systems are largely reactive. They depend on fixed infrastructure, human attention, and post-hoc analysis, which leads to delayed responses, blind spots, and inefficient use of resources.

I'm building Aarakshah to explore how autonomous systems can reason about space, movement and risk in real time, while adapting to information dynamically, rather than following rigid routes or rules.

I also hope to integrate this with my other drone-based projects, and hopefully scaling to city-wide use in the future.

The project is motivated by the need for scalable, proactive security systems that can operate continuously while still keeping humans in the loop through clear explanations and prioritized alerts. It also serves as an exploration of how AI can be used for situational analysis, decision support, and adaptive control in safety-critical environments.

How it works(High-Level)

At a conceptual level, Aarakshah operates as a closed-loop monitoring and control system built around spatial modeling, autonomous patrols, and AI-driven event reasoning.

• Spatial Modeling Layer: The monitored region is modeled as a graph derived from real-world geographic data, including roads, perimeters, access points, and zones of interest. This representation enables structured reasoning about coverage, movement, and risk distribution.

• Patrol Planning & Coverage: Autonomous aerial agents patrol the environment using efficient graph-based traversal strategies, including Eulerian path planning, to ensure complete and persistent area coverage with minimal redundancy.

• Sensing & Perception: Drones act as mobile sensing agents, collecting visual and contextual data through onboard sensors. Computer vision and signal analysis modules detect low-level anomalies such as unusual movement patterns, crowd formations, or interference.

• GenAI-based Situational Analysis: A GenAI-driven reasoning layer contextualizes detected events, assesses severity, and prioritizes risks. Rather than producing raw alerts, the system generates structured, explainable interpretations and recommended responses for operators.

• Adaptive Response & Control: Based on anomaly significance, patrol routes are dynamically adjusted to focus attention on high-risk areas while preserving overall surveillance coverage. Human operators remain in control, receiving prioritized insights rather than continuous raw feeds.

Data and Evaluation

The system uses OpenStreetMap data to construct realistic environment models and relies on synthetic and simulated datasets to train and test anomaly detection and patrol behavior. Evaluation is conducted through simulation-based experiments measuring:

• Patrol coverage efficiency

• Anomaly detection latency

• Adaptive response time

• Scalability with increasing area size and agent count