Sample paper
Word Count: approximately 2,000 words
Scope Statement
The report presents a requirements analysis specification for an autonomous drone navigation system's obstacle detection and path planning subsystem, describing its functional and temporal requirements using data flow diagrams, control flow diagrams, process specifications, and finite state machine-based control specifications. Key elements include sensor data processing for obstacle detection, real-time environment mapping, dynamic path planning, and reactive collision avoidance, providing a requirements baseline for subsystem design and implementation supporting safe autonomous flight in obstacle-rich environments.
Subsystem Description
The subsystem enables collision-free navigation to target locations through four functions: sensor data processing and fusion (ingesting camera, lidar, radar, and ultrasonic data to estimate obstacle position, velocity, and trajectory via computer vision and tracking algorithms); real-time environment mapping (integrating obstacle tracks into a dynamically updated occupancy map for situational awareness); path planning (computing minimum-length, obstacle-free trajectories respecting the drone's dynamics, continuously replanned as the map changes); and flight control command generation (guiding the drone along planned paths and triggering reactive evasive manoeuvres when collisions become imminent). Architecturally, detection and control run synchronously with sensor input, while mapping and planning are triggered asynchronously by detection and collision events, with minimum-latency detection and replanning essential to safety at maximum operating speed.
Refining the Description with Generative AI
The initial AI response correctly identified the four core functions and the latency-driven performance requirement but lacked implementation-level detail. The critique elaborated each function's specific algorithms and data sources, made explicit the dataflow and control architecture connecting processing stages, quantified latency and throughput constraints within the drone's operating envelope, and added emphasis on fail-safe and integrity requirements essential for flight safety assurance -- extensions the report frames as necessary to move from a high-level functional summary to a detailed requirements baseline.
Data Flow Diagrams
The context diagram shows real-time sensor feeds (cameras, lidar, radar, ultrasonics) as inputs, with obstacle positions and trajectories output to the drone's navigation and control subsystems. The process decomposition shows Obstacle Detection fusing multi-modal sensor data via computer vision and Kalman-filter tracking to produce consolidated obstacle tracks; Environment Mapping integrating these tracks into a probabilistic 3D occupancy grid; Path Planning searching the occupancy map with randomised planners (RRT, PRM) and graph search (A, D) to find safe, efficient trajectories, continuously updated as new map data arrives; and Motion Control translating planned paths into time-synchronised flight commands while monitoring obstacle tracks for imminent collisions.
Control Flow Diagrams
Obstacle detection processing is paced by sensor data arrival: Sensor Data Ready events trigger frame acquisition and tracking, with consolidated tracks pushed to mapping and planning on Obstacle Detected events. Mapping and planning proceed asynchronously, with significant map updates triggering a Map Updated event that initiates trajectory replanning, publishing revised paths to motion control on Path Generated events. The motion controller's nominal state executes the active flight plan, with Waypoint Reached events providing progress feedback; an Imminent Collision event interrupts path following at any point, transitioning the controller to a pre-scripted avoidance manoeuvre until an All Clear event returns it to waypoint following. Other events -- sensor failures, motor faults, low battery, lost link -- trigger a safe fallback state such as loiter or immediate landing.
Process Specifications
The Obstacle Detection process specification, triggered on each new sensor frame, extracts features from camera data, clusters lidar points, and filters radar returns, fusing these into consolidated detections. Each detection is matched against existing tracks (updating position and predicted trajectory) or spawns a new track; stale unmatched tracks beyond the configured tracking latency are pruned. Each tracked obstacle's projected trajectory is then checked against the drone's flight vector, emitting an imminent collision alert if intersection is unavoidable within the collision horizon, before publishing the updated track set downstream.
Control Specifications
The six-state model comprises IDLE (awaiting mission upload and launch), TAKEOFF (executing climb to altitude), WAYPOINTFOLLOW (navigating the planned mission), HAZARDAVOID (executing reactive avoidance), LANDING (autonomous landing at completion or fault), and FAULT (fault detection, isolation, and recovery). Transitions show that collision-imminent and fault-detected events can interrupt WAYPOINTFOLLOW at any point, that HAZARDAVOID returns to WAYPOINTFOLLOW once clear or escalates to FAULT if a fault also occurs, and that FAULT transitions to LANDING regardless of whether the fault is mitigated (precautionary landing) or critical (emergency landing/termination).
Conclusion and Next Steps
The analysis defines requirements for obstacle sensing, mapping, path planning, and collision avoidance, with the dataflow and control models capturing functional decomposition and behavioural states essential for autonomous flight, and PSPECs and CSPECs providing an executable model translatable into implementation logic and validation test cases. Next steps include defining process interface schemas, allocating functions to computing and networking resources, prototyping core algorithms in simulation, component-level PSPEC conformance testing, and integrated system testing validating flight safety across nominal and off-nominal scenarios.
References
Bieber, P., Boniol, F., Boyer, M., Noulard, E., & Pagetti, C. (2012). New challenges for future avionic architectures. Aerospace Lab, (4). Bieber et al., 2012 Fletcher, S., Norman, P., Galloway, S., & Burt, G. (2014). Impact of engine certification standards on the design requirements of More-Electric Engine electrical system architectures. SAE International Journal of Aerospace, 7(1), 24-34. Fletcher et al., 2014 Hobbs, A. (2010). Unmanned aircraft systems. In Human Factors in Aviation (pp. 505-531). Academic Press. Hobbs, 2010 Kochenderfer, M. J., & Chryssanthacopoulos, J. P. (2011). Robust airborne collision avoidance through dynamic programming. MIT Lincoln Laboratory Project Report ATC-371. Kochenderfer & Chryssanthacopoulos, 2011 Ray, A. (2004). Symbolic dynamic analysis of complex systems for anomaly detection. Signal Processing, 84(7), 1115-1130. Ray, 2004 Yin, J., Lawler, B., & Jin, H. (2017, September). Application of model based system engineering to IMA development activities. 2017 IEEE/AIAA 36th Digital Avionics Systems Conference (DASC) (pp. 1-7). IEEE. Yin et al., 2017