Constructing and Adapting Control Barrier Functions for Guaranteed Safe Control of Autonomous System
Abstract
In many naval applications, a human operator or autonomously-controlled system must have (a) theoretically guaranteed safe oper,ating regions and a safety override controller based on limited information about the system and environment, and (b) the ability to,rent theory and tools for safety verification have two major challenges. First, current methods struggle to be scalable while also b,eing applicable to general nonlinear system dynamics. Second, these safety guarantees are made on assumptions about the system and e,nvironment that may be invalidated during deployment. There is therefore a need to develop scalable approaches for safety analysis t,hat can adapt guarantees quickly based on changing assumptions. This proposal seeks to address these challenges by developing new th,ontrol barrier functions (CBFs). HJ reachability is the more general approach, and directly solves for the maximal safe region w,ithin which a system may operate. This approach is less scalable, though recent work from the PI and her colleagues has made advance,s scaling to higher-dimensional systems (6-12 state dimensions). A CBF generally has a more conservative safe region than HJ reachab,ility, but provides a useful quadratic program online controller that smoothly trades off between the desired performance control an,d the necessary safe control. For systems of specific forms, the CBF can be constructed analytically and can therefore be very scala,ble, but finding a valid CBF is in general challenging for nonlinear dynamic systems. There is active research on learning CBF appro,ximations for general system dynamics, and augmenting approximate CBFs with hand-tuned or learned "backup" controllers. Specific, thrusts for this proposal include: 1) Direct construction of CBFs. Recent work by the PI explores the connection between HJ rea,chability theory and CBF theory, and demonstrates how a CBF-like function with the largest possible safe operating region can be con,structed directly using HJ reachability tools. The PI will advance this work to identify the conditions under which one can directly, compute CBFs for general nonlinear dynamical systems. 2) Fast approximations with rigorous convergence to safety guarantees. Th,ere are many recent research efforts on efficiently approximating CBFs using data-driven approaches. Within HJ reachability there ar,e advances on converging to safety guarantees based on approximate initializations and decomposition. We propose that by using fast,CBF approximations as initializations for HJ reachability, scalable safety guarantees can be computed for general nonlinear dynamica,l systems. 3) Adapting safety online. There is active research on adapting control policiesfor CBFs based on information learned, on, safe learning to CBFs using these tools from both methods to update safety analyses efficiently with convergence guarantees. Moreov,er, we will explore how to adapt the safe set under large changes in system representations (e.g. changes in the state space) Co,mpletion of these research tasks will allow the community to perform safety verification for many real-world high-dimensional nonlin,ear systems, and to scalably update these safety analyses online based on learned information about the system and its environment.,This ability to use adaptive safe control will be a major asset to next-generation naval missions that must send autonomous systems,to perform tasks in environments with sparse, uncertain, and changing information that may affect the system s safety and performanc,e.(Approved for Public Release)
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 01, 2022
- Source ID
- N000142212292
Entities
People
- Sylvia Herbert
Organizations
- Office of Naval Research
- United States Navy
- University of California, San Diego