Title:
Differentiable and tolerant barrier states for improved exploration of safety-embedded differential dynamic programming with chance constraints
Differentiable and tolerant barrier states for improved exploration of safety-embedded differential dynamic programming with chance constraints
Author(s)
Kuperman, Joshua Ethan
Advisor(s)
Theodorou, Evangelos A.
Editor(s)
Collections
Supplementary to
Permanent Link
Abstract
A great challenge exists at the intersection of perception and controls – integrating the uncertainty present in perception-based state and obstacle estimation into safe control and trajectory optimization. First, we present the tolerant discrete barrier state (T-DBaS), a novel safety-embedding technique for trajectory optimization with enhanced exploratory capabilities. This approach generalizes the standard discrete barrier state (DBaS) method by accommodating temporary constraint violation during the optimization process while still approximating its safety guarantees. Towards applying T-DBaS to safety-critical au- tonomous robotics, we combine it with Differential Dynamic Programming (DDP), leading to the proposed safe trajectory optimization method T-DBaS-DDP, which inherits the con- vergence and scalability properties of the solver. Despite this, the tolerant barrier function parameters require tuning to reach peak performance for a wide array of constraints. To alleviate this requirement, we tune the T-DBaS parameters with the parameterized trajec- tory optimizer Pontryagin Differentiable Programming (PDP), proposing T-DBaS-PDP, an interpretable and generalizable solver for a variety of optimal control problems. In order to integrate perception uncertainty into safe optimal control, we learn the safety of the sys- tem via gaussian processes to create an interpretable, data-driven, and safety-guaranteeable framework. We implement this framework on differential drive and quadrotor dynamics and show its improvement over hand-tuned T-DBaS-DDP.
Sponsor
Date Issued
2023-05-10
Extent
Resource Type
Text
Resource Subtype
Thesis