Fast and Certifiable Trajectory Optimization

Shucheng Kang,Xiaoyang Xu,Jay Sarva,Ling Liang,Heng Yang

2024-09-03

Abstract:We propose semidefinite trajectory optimization (STROM), a framework that computes fast and certifiably optimal solutions for nonconvex trajectory optimization problems defined by polynomial objectives and constraints. STROM employs sparse second-order Lasserre's hierarchy to generate semidefinite program (SDP) relaxations of trajectory optimization. Different from existing tools (e.g., YALMIP and SOSTOOLS in Matlab), STROM generates chain-like multiple-block SDPs with only positive semidefinite (PSD) variables. Moreover, STROM does so two orders of magnitude faster. Underpinning STROM is cuADMM, the first ADMM-based SDP solver implemented in CUDA and runs in GPUs (with C/C++ extension). cuADMM builds upon the symmetric Gauss-Seidel ADMM algorithm and leverages GPU parallelization to speedup solving sparse linear systems and projecting onto PSD cones. In five trajectory optimization problems (inverted pendulum, cart-pole, vehicle landing, flying robot, and car back-in), cuADMM computes optimal trajectories (with certified suboptimality below 1%) in minutes (when other solvers take hours or run out of memory) and seconds (when others take minutes). Further, when warmstarted by data-driven initialization in the inverted pendulum problem, cuADMM delivers real-time performance: providing certifiably optimal trajectories in 0.66 seconds despite the SDP has 49,500 variables and 47,351 constraints.

Optimization and Control,Robotics

What problem does this paper attempt to address?

The paper aims to address the problem of non-convex trajectory optimization and proposes a new framework called STROM (Semidefinite Trajectory Optimization). Specifically, the study focuses on the following issues:

1. **Fast Solving**: Existing tools are slow in solving non-convex trajectory optimization problems, especially for large-scale problems. The proposed method can significantly improve solving speed, being two orders of magnitude faster than existing tools.
2. **Optimality Certification**: By using Semidefinite Relaxation to generate approximate solutions and provide optimality certification. STROM uses a sparse second-order Lasserre hierarchy to generate the semidefinite programming (SDP) relaxation form of the problem.
3. **GPU Acceleration**: To further improve solving speed, the paper proposes a GPU-accelerated algorithm called cuADMM. cuADMM is a variant of the Alternating Direction Method of Multipliers (ADMM) specifically optimized for GPUs.
4. **Real-time Performance**: In certain specific application scenarios (such as the inverted pendulum problem), STROM can achieve real-time optimal trajectory solving through data-driven initialization methods.

In summary, the paper is dedicated to developing a non-convex trajectory optimization method that is both fast and capable of certifying optimality, and can efficiently operate in practical applications.