Register for a range of sessions:
Virtual Testing for Robust Autonomous System Development
1:00 PM – 1:45 PM ET
Assuring the quality and performance of autonomous systems is a challenge due to the vast number of scenarios that can be encountered during operation. Verification and validation activities become increasing difficult as autonomy levels increase. Virtual testing provides methods to perform verification and validation tasks so that the autonomous technologies can be evaluated early and more efficiently leading to decreased risk and increased confidence during hardware testing. This session will discusses of different methodologies for virtual testing fit for various scenarios and how they can prove system robustness while accelerating development. Virtual testing methodologies range from simple simulations of algorithms to closed-loop simulation with sensor models and visualization. Methods of managing and accelerating virtual testing through automation will also be explored. These topics will address the challenges in assuring that new autonomous systems are developed correctly and efficiently.
COVID shutdowns accelerated use-cases for drones, and with more drones in the skies, come more exposed vulnerabilities. Drones have always been able to go where you cannot go by yourself. This was the truth before COVID, and it is accentuated even more now. Organizations using drone must be aware of how to differentiate the drones that are a part of their program, ensure compliance with local, state, or federal laws, and expose an unauthorized or hostile drone in their area. With the growth of the drone market and the increased savviness of drone pilots, government and private organizations need to shift their mindset from focusing purely on the counter-drone technology, to what they truly need to achieve complete airspace security. Cities of the future will require complete airspace awareness on their drone activity. As unmanned traffic management systems are being developed and deployed globally, AUVSI members will be the first to understand the true nature of drone activity. This presentation will discuss the evolution of counter-drone technology, current use applications for cities, enterprises, and government organizations, and how AUVSI members can begin to cooperate with counter-drone technology platforms and integrate airspace security analytics into their existing drone infrastructure.
Deep Teaching: A Scalable AI Approach to Autonomous Driving
2:00 PM – 2:45 PM ET
Today’s L4 autonomous vehicle deployments are limited by the tremendous cost and time of development of the safety critical AI software which is required to address the enormous tail end of corner case scenarios. We will cover an emerging AI technology called Deep Teaching, which reduces the cost and time of development of autonomous driving systems by allowing large scale training of neural networks without the bottleneck of human annotation and by tackling training in the small data regime. We’ll conclude with demonstrations of the results of Deep Teaching, including state of the art performance on today’s autonomous driving benchmarks.
Building Trust in Autonomous Sensors
2:00 PM – 2:45 PM ET
Edge computing and efficient artificial intelligence (AI) algorithms are revolutionizing the speed with which relevant, actionable results can be generated from sensor data. Users no longer have to review sensor data by hand or wait for data to be transmitted to powerful remote servers for processing. Instead, small and powerful processors are built into sensors and produce valuable intelligence within the sensor enclosure. For example, a smart camera can detect obstacles in a vehicle’s path or monitor a controlled area to alert security to trespassers, all without any additional computer resources. These smart sensors generate results autonomously, but users, whether in the loop, on the loop, or outside the loop, still need to develop trust before these autonomous sensors can be effectively deployed. Our paper presents several techniques for fostering trust in autonomous sensors. These techniques fall into three main categories. First, we involve users in the interface design and development process. Second, we conduct evaluations with the goal of building system trust. Third, we incorporate explainable AI software to illuminate the inner workings of the usually opaque deep learning systems used for sensor processing. These techniques build trust in AI-enabled sensors and facilitate their adoption.
The Multicore Challenge in Assured Autonomy
3:00 PM – 3:45 PM ET
Assured autonomy requires that the autonomous system behaves correctly for all scenarios under which it operates. The software in an autonomous system must compute the right result and do so at the right time, and ideally one would like to prove correctness of an autonomous system before putting it into operation. This issue is widely recognized by academic researchers and practitioners in autonomy and solutions have been presented, including some of our own solutions presented at AUVSI XPONENTIAL 2019 and 2020. Autonomous systems increasingly use multicore processors and these bring additional challenges related to correct timing however. These challenges, along with solutions, is presented in this paper. In particular, timing of software executing on multicore processors depends on hardware resources (e.g., cache memories, memory controllers, buses) shared between processor cores but often these resources are un-documented. Thus, one often faces the situation that one wants to formally verify a property (correct timing) of a system although one does not have a full description of the system. We discuss this issue; we present a previously known state-of-the-art offline verification technique and also present a new preliminary idea for run-time enforcement of software for this situation.
Advancing Autonomy through Platform Integration
3:00 PM – 3:45 PM ET
As the U.S. government moves toward the development and fielding of fully autonomous systems, the integration of elements that support autonomous functionality becomes increasingly important. During this panel discussion, moderated by Peraton, attendees will hear from leaders at the resource sponsor, program office and operational level, as they discuss the most critical topics surrounding platform hardware and software integration, necessary to achieve assured autonomy. Discussion will include sensor suite and onboard computing topics such as accurate position location; object and traffic detection, identification, tracking, and avoidance; as well as the effects of natural and man-made environmental conditions. Panelists will also explore critical software integration topics, such as artificial intelligence and machine learning capability and speed; autonomous route planning; hierarchical system of avoidance and responsiveness; autonomous rerouting; anti-spoofing; and sensor plug-and-play capabilities. The information presented in this panel is critical to anyone seeking to understand how the Department of Defense is approaching autonomy integration into current and future programs.
This paper details the buildup and execution of a laboratory demonstration simulating the high-endurance flight of a VTOL fixed-wing platform. High-endurance flight is achieved via hybrid-electric propulsion. The lab demonstration occurs in the Kent State University Electric Vertical Takeoff and Landing (EVTOL) Propulsion Lab. This paper discusses the purpose and objectives of the research, a description of the eVTOL Propulsion Lab and its capabilities, the mission profiles and scale of power requirements, a generic overview of the hybrid-electric system, lab demonstration performance, and analysis of the results.