CuPIQP: A Batched and Differentiable QP Solver on GPUs for Robot Learning and Control
Organizations: Automatic Control Laboratory, EPFL, Switzerland. · Accessible and Accelerated Robotics Lab, Dartmouth College, USA. · Institute for Transport Planning and Systems, ETH Zürich, Switzerland.
Abstract
Quadratic programs (QPs) are fundamental building blocks of many robotic tasks such as inverse kinematics, whole-body control, and model predictive control. Extending these methods to large-scale robot learning and control requires solving many QPs concurrently and, for end-to-end learning, differentiating through their solutions. Efficient batched optimization is therefore essential to bringing established model-based methods into GPU-based robotics pipelines. We present cuPIQP, a batched and differentiable QP solver on GPUs based on a proximal interior-point method, with unified support for dense, sparse, and multistage problems. The solver exploits parallelism across independent QPs and their internal structure while preserving per-problem numerical updates and termination decisions. Benchmarks with up to 16384 QPs per batch demonstrate solve-time speedups of up to over Moreau and over eight-core PIQP. In an end-to-end learning task, cuPIQP accelerates training by over QPAX and over Moreau. By reducing the computational cost, cuPIQP provides efficient, reusable infrastructure for integrating model-based optimization into large-scale robot learning and control.
Figures & tables
| Solver | QP fwd. | QP bwd. | Other | Total | QP fwd. | QP bwd. | Other | Total | QP fwd. | QP bwd. | Other | Total |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| cuPIQP -dense | 1.37 (80%) | 0.09 (5%) | 0.26 (15%) | 1.72 | 3.83 (88%) | 0.25 (6%) | 0.27 (6%) | 4.36 | 14.26 (92%) | 0.92 (6%) | 0.29 (2%) | 15.47 |
| cuPIQP -multistage | 0.43 (60%) | 0.04 (5%) | 0.25 (35%) | 0.72 | 0.91 (74%) | 0.07 (6%) | 0.25 (20%) | 1.24 | 2.88 (87%) | 0.18 (5%) | 0.26 (8%) | 3.31 |
| cuPIQP -sparse | 0.72 (70%) | 0.06 (5%) | 0.25 (24%) | 1.03 | 1.74 (82%) | 0.12 (6%) | 0.26 (12%) | 2.12 | 5.28 (90%) | 0.30 (5%) | 0.27 (5%) | 5.85 |
| Moreau | 2.89 (82%) | 0.38 (11%) | 0.26 (7%) | 3.53 | 8.79 (86%) | 1.16 (11%) | 0.27 (3%) | 10.21 | 35.18 (89%) | 4.20 (11%) | 0.28 (1%) | 39.66 |
| QPAX | 2.27 (85%) | 0.12 (5%) | 0.27 (10%) | 2.66 | 5.14 (90%) | 0.24 (4%) | 0.34 (6%) | 5.72 | 18.03 (94%) | 0.74 (4%) | 0.34 (2%) | 19.11 |
Quadratic Programming
Quadratic programming optimizes quadratic objectives under constraints, supporting applications from robotics and control to finance and machine learning.
See all 24 papers