Unmanned aerial vehicles (UAVs) are usually interfaced with a remote operator via a satellite link. This arrangement is inherently high-latency and has delays when responding to operator commands. To mitigate this, a real-time onboard artificial intelligence is added to analyze the sensory input and autonomously control the aircraft.
We propose a project that would improve on our prior research in this area, including making improvements to a functioning software system. We have proof of concept demonstration (shown below), worked out applied math, parameters, and have purpose built a high performance software library that utilizes parallel multi-core execution.
Our further development will streamline easy-to-use scripts for setting up and updating the mission plans. The system was built on top of a custom rendering engine, utilizing open source and cross-platform libraries. It will be provided for visualizing the decision making by the AI system, and for simulating various mission scenarios.
Key features of the project outcome:
1. Real-time stateless situational awareness (AI will produce instantaneous control commands).
2. Instant availability of manual controls for operator to override/resume autonomy.
3. Static and dynamic obstacles.
4. Human-readable and easily interpretable control instructions to instantly review and log.
5. Supports multiple presets for upload in real-time.
6. Presets can be reviewed/simulated at a high fast-forward speed, scrubbed and analyzed on varying input data, then released for execution.
7. Presets feature obstacles/threats to avoid, and targets to reach. Conflicting goals are combined in an efficient scenario for optimal mission control.
8. Multiple waypoints can be scripted to dynamically trigger updates to the autonomous tactical scenario. It can explicitly request for manual input at waypoints.
9. Sensory input is non-homogenous by design. The control system is upgradable, will accept and utilize novel sensors.
10. Works on off-the-shelf 10 year old mobile processors, which means it is comparable with the current military-grade computing.
11. Autonomous maneuvering at low altitude and when other air traffic is present. Dynamically prioritizes high-priority tactical goals over a preset mission plan.
12. Easy to test in a virtual environment and maintain. This includes observing FAA exclusion no-fly zones.
Possible tactical scenarios include:
1. Low flying to avoid detection by anti-aircraft defense systems.
2. Navigating a challenging terrain, including high-rise structures and urban environments.
3. Navigating in open seas, where trajectory might intersect busy ship traffic, oil rigs etc.
4. Avoiding unexpected moving and flying obstacles.
Real-time demonstration.
Note the use of spheres to represent avoidance areas of simulated moving obstacles. Mission waypoints are represented by vertical cylinders.
About myself.
A graduate student at the University of Washington studying Applied Mathematics, will pursue a post doctorate degree. Having graduated with prior masters degrees in Computer Science and Electronics Engineering. My research is predominantly focused in applied mathematics, computational fluid dynamics, and machine learning.
No comments:
Post a Comment