CAREER: Towards Environment-Aware Adaptive Safety for Learning-Enabled Multiagent Systems with Application to Target Drone Capturing

Overview Current anti-drone technologies are often expensive and ineffective. The PI proposes a concept using a team of unmanned aerial vehicles (UAVs) with a capture net called “MUCH-Net”, which stands for “Multi-UAV Drone Catch Net”. This ?ve-year career-development plan is to develop an integrated research, education, and outreach program on designing learning-based cooperative control frameworks for multiagent systems with environment-aware adaptive safety requirements, in the context of “MUCH-NET” operation. The objectives and approaches are: • To investigate frameworks to address environment-aware adaptive safety. “Environment-aware” means safety requirements should adapt to the environment. “Adaptive” means safety should adapt to the presence of external agents/target. A novel “Composite Barrier Function” (CBF) structure, which integrates with an environment parameter learning update law, is proposed. This CBF structure also incorporates adaptive neural network learning schemes to learn the target velocity, system dynamics, and environment disturbances. • To address safety requirement con?icts. To extend feasible solution space, potential initial state safety con?icts are addressed by integrating initial state and virtual barriers into CBFs. Indicator-based CBFs are designed to resolve potential con?icts in different safety sets. • To stimulate learning and promote engineers. The PI will create cooperative learning environment with K-12 students and teachers, introduce exploratory course assignments, and integrate research with curriculum development. • To work closely with industries to promote research development and technology transfer. This objective will be achieved by collaborating with several local and national industry partners. Intellectual Merit Existing safety-critical control algorithms only address constant or time-varying safety sets, which cannot dynamically adapt to the environment and external agents. These algorithms also cannot accommodate potential safety con?icts by the initial system state, and cannot resolve potentially con?icting safety requirements. The PI proposes a new environment-aware adaptive safety analysis. Speci?cally, the novel CBFs are dynamic barrier functions that integrate with a learning scheme to update a multi-dimensional environment parameter, so that the CBFs can respond to the environment and external agents/target. The CBFs also incorporate initial system state and virtual barriers to resolve any initial state safety con?icts. To resolve potential con?icts between different environment-aware adaptive safety requirements, indicator functions are incorporated to modify the less critical (“soft”) safety sets, in order to guarantee the more critical (“hard”) safety requirements. The PI is uniquely quali?ed for this project by virtue of his expertise in adaptive and learning control, safety-critical control, and control of nonlinear multiagent systems. Broader Impacts The proposed cooperative control architecture will have wide ranges of applications in security management, transportation, and other critical areas. In collaboration with our Department of STEM Education in the College of Education (see Letters of Collaboration), a concept design contest, a robot capture game competition, and a robotic program for K-12 teachers, which involve women, minorities, and students from under-represented groups, will be proposed based on the research in this CAREER project. The PI will promote exploratory learning assignments in undergraduate teaching and develop a new graduate course that integrates with this project. Collaborations with industry partners including Aviation Safety Resources (ASR), Aerial Vehicle Automation (AVA), and Cincinnati/Northern Kentucky International Airport (CVG) will facilitate research development and technology transfer (see Letters of Collaboration).