ICLR 2026 Orals

Differentiable Model Predictive Control on the GPU

Emre Adabag, Marcus Greiff, John Subosits, Thomas Jonathan Lew

Reinforcement Learning & Agents Sat, Apr 25 · 3:27 PM–3:37 PM · 202 A/B Avg rating: 7.33 (6–8)

Abstract

Differentiable model predictive control (MPC) offers a powerful framework for combining learning and control. However, its adoption has been limited by the inherently sequential nature of traditional optimization algorithms, which are challenging to parallelize on modern computing hardware like GPUs. In this work, we tackle this bottleneck by introducing a GPU-accelerated differentiable optimization tool for MPC. This solver leverages sequential quadratic programming and a custom preconditioned conjugate gradient (PCG) routine with tridiagonal preconditioning to exploit the problem's structure and enable efficient parallelization. We demonstrate substantial speedups over CPU- and GPU-based baselines, significantly improving upon state-of-the-art training times on benchmark reinforcement learning and imitation learning tasks. Finally, we showcase the method on the challenging task of reinforcement learning for driving at the limits of handling, where it enables robust drifting of a Toyota Supra through water puddles.

One-sentence summary·Auto-generated by claude-haiku-4-5-20251001(?)

DiffMPC provides GPU-accelerated differentiable MPC solver leveraging problem structure for efficient parallelization.

Contributions·Auto-generated by claude-haiku-4-5-20251001(?)
  • GPU-accelerated differentiable optimization tool for model predictive control
  • Sequential quadratic programming with custom preconditioned conjugate gradient routine
  • Exploits time-induced sparsity in optimal control problems for efficient parallelization
  • Substantial speedups over CPU and GPU baselines on RL and imitation learning benchmarks
Methods used·Auto-generated by claude-haiku-4-5-20251001(?)
  • Model predictive control
  • Differentiable optimization
  • Sequential quadratic programming
  • Preconditioned conjugate gradient
Limitations (author-stated)·Auto-generated by claude-haiku-4-5-20251001(?)
  • Inequality constraints require penalization in cost or control bounds in dynamics
    from the paper
  • Differentiating through inequality constraints remains challenging with gradient discontinuities
    from the paper
  • Runs slower on CPU than GPU due to JAX implementation
    from the paper
  • Does not explicitly support tuning solver hyperparameters
    from the paper
  • Poor initial guesses can result in solver divergence and hindered downstream training
    from the paper
Future work (author-stated)·Auto-generated by claude-haiku-4-5-20251001(?)
  • Handle inequality constraints via augmented Lagrangian or interior-point methods
    from the paper
  • Rewrite solver in C/C++ for CPU performance improvements
    from the paper
  • Develop robust initialization methods for differentiable optimization pipelines
    from the paper

Author keywords

  • differentiable optimization
  • model predictive control
  • optimal control
  • gpu-accelerated optimization
  • reinforcement learning
  • imitation learning
  • robotics

Related orals

Something off? Let us know →