Expandable Variable-Autonomy Architecture (EVAA)

Effective multi-level autonomous piloting systems require integration with safety-critical functions. The Expandable Variable-Autonomy Architecture (EVAA) project seeks to develop a hierarchal autonomous system framework that will depend on deterministic systems with higher authority to protect against catastrophic piloting faults and allow a lower level certification for the machine learning sub-systems. The multi-layered approach provides the framework for analytical systems that can learn, predict, and adapt to both routine and emergency situations. 

The objective of the project is to develop an autonomous piloting system based on analytical and learning algorithms that are capable of making effective decisions, in both nominal and potentially catastrophic situations. This will develop a safety critical framework for certification of complex autonomous systems where a small but sufficient number of levels. The system will be integrated with a certified safety critical decision makers (such as vehicle health monitoring, collision avoidance, loss of control avoidance and restricts commands of higher level critical decision makers not certified to level A software. The project will integrate these systems onto a quad-rotor micro-UAV for inexpensive and quick flight testing of concepts and develop customized, low power hardware to house the control and decision making algorithms.

ASSUMPTIONS AND LIMITATIONS: The purpose of this CIF project is not to develop a full scale aircraft capable of these types of advancements, but only to develop a piloting system which make them possible. Initially, decisions associated with “where to fly” will be focused on and integrated into the algorithms. For this slice of the pie, the system will be required to navigate a potentially changing dense urban landscape. Routes will be planned based on time, distance, and potential risk. Additionally, terrain and obstacle avoidance algorithms will restrict these activities based on preloaded obstacle and terrain maps. Additionally, off nominal conditions such as loss of motor or other non-pre-programmed events will cause the aircraft to select landing or crashing locations based on population density maps, location of buildings, and other information. A hangar or small area will be turned into the urban city-center mockup with maps created of the mockup to facilitate flight test of concepts.

Work to date: The hierarchical decision chain and framework, hardware, and embedded processing related to ground collision avoidance is in place for a sub-scale platform. Flight tests on a quad-rotor model helicopter demonstrated successful limitation of flight decisions when facing imminent ground collision. 

Looking ahead: The team is developing a full set of safety-critical functions for the sub-scale platforms and working to scale up to larger UAVs. 

Partners: University of California at Berkeley and Stanford University are developing algorithms, and the FAA is participating in the certification process. 

Benefits 

• Increases safety: Integration of safety-critical functions improves outcomes in emergency situations. 
• Certifiable: Removal of safety-critical functions from the autonomous control enables adaptable processes to be certified to a lower level. 

Applications 

• UAVs and unmanned submersibles 
• Autonomous rail transport 
• Deep space exploration 
• Driverless vehicles