Learning-Based Approximate Nonlinear Model Predictive Control Motion Cueing

Arango, Camilo Gonzalez; Asadi, Houshyar; Qazani, Mohammad Reza Chalak; Lim, Chee Peng · 2025 · ArXiv.org

DOI: 10.48550/arxiv.2504.00469

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the computational bottleneck of Nonlinear Model Predictive Control (NMPC) in Motion Cueing Algorithms (MCAs) for serial robot-based motion simulators. While NMPC offers superior handling of nonlinear constraints and kinematic modeling compared to traditional filter-based methods, its high computational cost prevents real-time operation at high control rates. The authors propose a novel learning-based MCA that merges the accuracy of NMPC with the efficiency of machine learning by shifting the computational burden to an offline training phase. This approach utilizes a Differentiable Predictive Control (DPC) framework, enabling real-time inference while maintaining rigorous adherence to joint acceleration, velocity, and position limits. The method employs a nonlinear joint-space plant model for a KUKA KR500 robot arm simulator. The DPC architecture consists of two components: a "plant" model that predicts system dynamics and a "policy" network that determines control inputs. The authors evaluated three state transition models (Neural State Space Model, Neural Ordinary Differential Equations, and an exact analytical implementation) and three output prediction models (NSSM, Latent Space Model, and a Mixed Analytical Model). The policy network is trained to mimic NMPC behavior using an MPC-inspired loss function that penalizes tracking error, control effort, and constraint violations. Training occurs in two stages: first, the plant model is trained in open-loop mode to learn system dynamics; second, the policy is trained in a closed-loop setup using the pre-trained plant to optimize regulation performance over prediction horizons. Simulation experiments demonstrated that the proposed algorithm performs on par with state-of-the-art NMPC baselines in terms of motion cueing quality, as measured by Root Mean Square Error (RMSE) and correlation coefficients with reference signals. Crucially, the learning-based approach was on average 400 times faster than the NMPC baseline, enabling real-time operation at high control rates. The algorithm also exhibited strong generalization capabilities, successfully handling unseen operating conditions, including motion cueing scenarios on different vehicles and real-time physics-based simulations. The significance of this work lies in establishing a new paradigm for motion cueing that overcomes the trade-off between computational efficiency and control accuracy. By approximating NMPC with a learned policy, the method retains the ability to handle complex nonlinear constraints and kinematic models without the associated online optimization costs. This enables high-fidelity, realistic driving experiences in motion simulators that were previously computationally prohibitive, marking a significant advancement in the application of machine learning to control systems for serial robot platforms.

Key finding

The proposed learning-based motion cueing algorithm performs on par with NMPC in cueing quality but is on average 400 times faster, enabling real-time operation.

Methodology

simulation_modeling

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via author_sweep_intake on 2026-05-28.

StageOutcomeToolModelPromptAttemptsCompleted
discover success author_sweep 2 2026-05-28
archive success canonical_url 1 2026-06-04
extract success cached 3 2026-06-10
clean success clean 1 2026-06-04
chunk success chunk 1 2026-06-04
embed success embed Qwen/Qwen3-Embedding-8B 1 2026-06-04
enrich success 1 2026-05-28
promote success 1 2026-06-04
summarize success llm qwen3.6-27b-prismaquant summ-v5 2 2026-06-10
tag success vector_similarity 15 2026-06-11
verify success 2 2026-06-10

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).