Tag Archives: Parallelization

On the use of GPUs for parallelization of MPCs through the parallelization of symbolic mathematical expressions

January 9, 2025 16:29 ,

S. H. Jeon, S. Hong, H. J. Lee, C. Khazoom and S. Kim, CusADi: A GPU Parallelization Framework for Symbolic Expressions and Optimal Control, IEEE Robotics and Automation Letters, vol. 10, no. 2, pp. 899-906, Feb. 2025, DOI: 10.1109/LRA.2024.3512254.

The parallelism afforded by GPUs presents significant advantages in training controllers through reinforcement learning (RL). However, integrating model-based optimization into this process remains challenging due to the complexity of formulating and solving optimization problems across thousands of instances. In this work, we present CusADi , an extension of the casadi symbolic framework to support the parallelization of arbitrary closed-form expressions on GPUs with CUDA . We also formulate a closed-form approximation for solving general optimal control problems, enabling large-scale parallelization and evaluation of MPC controllers. Our results show a ten-fold speedup relative to similar MPC implementation on the CPU, and we demonstrate the use of CusADi for various applications, including parallel simulation, parameter sweeps, and policy training.

Posted in: Computer science , Tagged: , ,

Study of how a complex motion planning problem solved through RRT can benefit from parallelization

November 12, 2015 19:59 ,

Brian W. Satzinger, Chelsea Lau, Marten Byl, Katie Byl, Tractable locomotion planning for RoboSimian, The International Journal of Robotics Research November 2015 vol. 34 no. 13 1541-1558, DOI: 10.1177/0278364915584947.

This paper investigates practical solutions for low-bandwidth, teleoperated mobility for RoboSimian in complex environments. Locomotion planning for this robot is challenging due to kinematic redundancy. We present an end-to-end planning method that exploits a reduced-dimension rapidly-exploring random tree search, constraining a subset of limbs to an inverse kinematics table. Then, we evaluate the performance of this approach through simulations in randomized environments and in the style of the Defense Advanced Research Projects Agency Robotics Challenges terrain both in simulation and with hardware.
We also illustrate the importance of allowing for significant body motion during swing leg motions on extreme terrain and quantify the trade-offs between computation time and execution time, subject to velocity and acceleration limits of the joints. These results lead us to hypothesize that appropriate statistical “investment” of parallel computing resources between competing formulations or flavors of random planning algorithms can improve motion planning performance significantly. Motivated by the need to improve the speed of limbed mobility for the Defense Advanced Research Projects Agency Robotics Challenge, we introduce one formulation of this resource allocation problem as a toy example and discuss advantages and implications of such trajectory planning for tractable locomotion on complex terrain.

Posted in: Robot motion planning , Tagged: ,