MSC: Multi-Stage Caster-aware Control Framework with Uncertainty-Aware NMPC and RL-based Execution Layer

Jinho Park; Jeongtae Lee; Jeonghyeok Lim; Keun Ha Choi; Kyung-Soo Kim

cs.ROOct 12, 2026

MSC: Multi-Stage Caster-aware Control Framework with Uncertainty-Aware NMPC and RL-based Execution Layer

Authors: Jinho Park, Jeongtae Lee, Jeonghyeok Lim, Keun Ha Choi, Kyung-Soo Kim

Organizations: Department of Mechanical Engineering, KAIST, Daejeon, Republic of Korea

Abstract

Freely swiveling casters are widely adopted in industrial AMRs/SDVs for their mechanical simplicity and load capacity. However, caster misalignment during low-speed or high-curvature maneuvers induces strong contact nonlinearities, widening the planning-execution gap between commanded $(v, ω)$ and realized motion. Existing caster-aware strategies are highly sensitive to state-estimation reliability, causing unstable execution in uncertain transition regimes. To address this, we present Multi-Stage Caster-aware Control (MSC), a planning-execution integrated hierarchical framework featuring two core mechanisms. First, an uncertainty-aware NMPC (CU-MPC) dynamically schedules caster-aware costs and input regularization using predictive uncertainty to suppress overly aggressive maneuvers. Second, a reinforcement learning-based execution layer estimates and predicts caster states under partial observability, transforming high-level velocity commands into physically executable wheel speeds adapted to current contact conditions. Validations in Isaac Sim and real robots demonstrate that MSC offers a clear comparative advantage over existing decoupled compensation approaches by significantly improving caster-state estimation accuracy, minimizing misalignment, and enhancing tracking robustness in caster-sensitive maneuvers.

Figures & tables

Fig. 1: Caster misalignment and its navigation impact. A mismatch between the commanded motion direction and the caster yaw angle direction generates a lateral scrub contact force and a resistive caster-induced reaction torque, contributing to a planning–execution gap between planned $(v,\omega)$ and physically realizable motions.

Fig. 2: Overall framework of MSC. The caster-uncertainty-aware CU-MPC generates local velocity references. An RL-based Execution Layer, trained with privileged information in simulation, estimates/predicts caster states with uncertainty and transforms the high-level $(v,\omega)$ commands into dynamically feasible wheel commands. This hierarchical integration mitigates the planning–execution gap arising from caster-induced contact nonlinearities.

Fig. 3: Trajectory tracking comparison against baselines. Proposed Method improves tracking smoothness and accuracy over B1/B2, particularly under caster-induced contact nonlinearities.

Parameter	Symbol	Value
MPC (CA-MPC / CU-MPC)
NMPC horizon	$N$	63
Sampling time	$\Delta t$	0.05
State weight (tracking)	$\mathbf{Q}$	1.0
Input weight (base)	$\mathbf{R}$	0.1
Caster-aware weight	$w_{\mathrm{cw}}^{0}$	1.0

TABLE I: Key parameters used in simulation, learning, and MPC.

Fig. 4: Caster estimation/prediction and uncertainty-aware scheduling (in-place low-speed rotation). (a) Left caster angle comparison: GT, kinematic estimator, learned estimator, and learned predictor with the prediction uncertainty band. (b) CU-MPC weight scheduling driven by the predicted risk/uncertainty $\rho$ : caster-aware weight $w_{\mathrm{cw}}$ and control regularization weight $w_{\mathrm{ctrl}}$ .

Fig. 5: Real-world experimental results (C-path). (a) Executed trajectories for the compared methods in the mapped environment. (b) Caster angle estimation compared against encoder ground truth (GT), highlighting the phase lag of the kinematics-based estimator and the improved agreement of the proposed learning-based estimator. (c) Linear velocity tracking error $e_{v}$ over time. (d) Angular velocity tracking error $e_{\omega}$ over time.

Metric	M-path			S-path			C-path
	B1	B2	Ours	B1	B2	Ours	B1	B2	Ours
RMSE $[d]$ (m)	0.154	0.080	0.094	0.213	0.211	0.092	0.213	0.058	0.043
RMSE $[e_{\psi}]$ (rad)	0.532	0.472	0.533	0.173	0.474	0.176	0.127	0.184	0.129
RMSE $[v]$ (m/s)	0.007	0.017	0.019	0.006	0.020	0.013	0.006	0.013	0.011
RMSE $[\omega]$ (rad/s)	0.069	0.045	0.029	0.074	0.057	0.032	0.068	0.027	0.024
RMSE $[e_{\hat{\varphi}}]$ (rad)	0.301	0.276	0.029	0.289	0.425	0.065	0.390	0.712	0.064

TABLE II: Simulation results in Isaac Sim (M/S/C paths).

Fig. 6: Real-world experimental setup. (a) The AMR hardware platform used in the experiments. (b) ROS 2-based navigation stack visualization during execution (planning/localization and trajectory tracking).

Metric	B1	B2	Ours
RMSE $[d]$ (m)	0.1990	0.1171	0.1163
RMSE $[e_{\psi}]$ (rad)	0.3695	0.2113	0.2195
RMSE $[v]$ (m/s)	0.0083	0.0097	0.0059
RMSE $[\omega]$ (rad/s)	0.0902	0.0719	0.0645
RMSE $[e_{\hat{\varphi}}]$ (rad)	0.3450	0.2174	0.1005
RMSE $[e_{\mathrm{mis}}]$ (rad)	0.1880	0.1006	0.0849

TABLE III: Real-robot experiment results (C-path).

Robot Trajectory Tracking

Robot trajectory tracking develops methods that make robots follow desired paths or motions accurately despite disturbances, uncertainty, and physical constraints.

See all 137 papers

Lightweight Model Predictive Control for Spacecraft Rendezvous Attitude Synchronization
Mar 19, 2026Peter Stadler, Alexander Meinert, Niklas Baldauf +1Spacecraft Attitude Determination Model Predictive Control
Space Applications Group, e:fs TechHub GmbH, 85080 Gaimersheim, Germany · Faculty of Electrical Engineering and Information Technology, Ingolstadt University of Applied Sciences, Germany
Quantile-Based Command-Reserve MPC for Current-Command Gimbal Pointing
Oct 26, 2026Yaokun Lu, Yifan Li, Enmei Wang +3Model Predictive Control Disturbance Rejection
School of Automation Science and Electrical Engineering, Beihang University, Beijing, 100191, China · Hangzhou Innovation Institute of Beihang University, Hangzhou, 311115, China · School of Instrumentation Science and Optoelectronic Engineering, Beihang University, Beijing, 100191, China
MR-MPCC: Model Predictive Contouring Control for Robotic Manipulator Tracking under Moving Reference Frames
Oct 15, 2026Chonghao Cheng, Tiancheng Li, Jonathan Woolfrey +2Robotic Control Robot Trajectory Tracking
Robotics Institute, University of Technology Sydney, Sydney, NSW, Australia. · School of Informatics, University of Edinburgh, Edinburgh, UK.

Mar 19, 2026J/K move · Enter open · S save

Lightweight Model Predictive Control for Spacecraft Rendezvous Attitude Synchronization

Peter Stadler, Alexander Meinert, Niklas Baldauf, Alen Turnwald

Space Applications Group, e:fs TechHub GmbH, 85080 Gaimersheim, Germany · Faculty of Electrical Engineering and Information Technology, Ingolstadt University of Applied Sciences, Germany

Spacecraft Attitude Determination Model Predictive Control Feedback Control Spacecraft Rendezvous and Proximity Operations

Open paper AskPDF arXiv

This work introduces two lightweight model predictive control (MPC) approaches for attitude tracking with reaction wheels during spacecraft rendezvous synchronization. Both approaches are based on a novel attitude deviation formulation, which enables the use of inherently linear constraints on angular velocity. We develop a single-loop and a dual-loop MPC; the latter embeds a stabilizing feedback controller within the inner loop, yielding a linear time-invariant system. Both controllers are implemented with CasADi - including automatic code generation - evaluated across various solvers, and validated within the Basilisk astrodynamics simulation framework. The experimental results demonstrate improved tracking accuracy alongside reductions in computational effort and memory consumption. Finally, embedded delivery to an ARM Cortex-M7 - representative of commercial off-the-shelf devices used in New Space platforms - confirms the real-time feasibility of these approaches and highlights their suitability for onboard attitude control in resource-constrained spacecraft rendezvous missions.

Abstract

Robot Trajectory Tracking

Explore similar work

Lightweight Model Predictive Control for Spacecraft Rendezvous Attitude Synchronization

Quantile-Based Command-Reserve MPC for Current-Command Gimbal Pointing

MR-MPCC: Model Predictive Contouring Control for Robotic Manipulator Tracking under Moving Reference Frames