Robust Geometric Data-Driven Tracking for a Feedback-Linearised 2R Robot Arm

Pranav Gupta, Deepan Muthirayan, Ravi Banavar

5 min read•Nov 18, 2026

Predictive control chooses future inputs by predicting the system’s response, applies the first input, and repeats the optimisation after each new measurement. Model predictive control predicts with a mathematical model. Data-driven predictive control predicts from recorded data instead, as in data-enabled predictive control, which represents trajectories as linear combinations of recorded input–output data. The Fundamental Lemma justifies this representation for controllable linear time-invariant systems with noiseless, sufficiently exciting data.

Nonlinear systems with uncertain parameters challenge both approaches: inaccurate mechanical models yield biased predictions, while linear combinations of measured trajectories approximate nonlinear dynamics only locally. Model knowledge can reduce the data burden. For fully actuated manipulators, feedback linearisation uses a mechanical model to map virtual acceleration commands to joint torques. With exact parameters and states, each joint becomes a double integrator from virtual input to position. Parameter, state-estimation, and actuator errors produce residual dynamics that can be learned from virtual-input and position recordings.

Geometric robust data-driven predictive control fits a low-dimensional trajectory subspace and handles fitting errors using an open chordal-distance ball of same-dimensional subspaces on the Grassmannian. It minimises worst-case weighted squared deviation from a target combining measured history and a desired future trajectory. The mechanical model supplies the feedback-linearising torque map; data-driven predictive control predicts the virtual-input-to-position response. This study tests whether learning this response improves prediction and tracking beyond the known double-integrator model.

Planar 2R Manipulator Control

The mechanical model feedback-linearises the arm, and geometric data-driven predictive control predicts the resulting virtual-input–position response.

Offline Data and Reference Initialisation

A PD tracking controller with acceleration feedforward and random excitation generates offline data. Its sinusoidal reference varies in centre, amplitude, phase, and period between episodes.

The reference is a sinusoid whose centre, amplitude, phase, and period vary between episodes.

Simulation Studies

The study examines prediction accuracy (A), transfer of predictive tuning (B), and cyclic tracking (C). Controllers receive noisy positions; true positions are used only to evaluate errors.

Prediction Accuracy After Feedback Linearisation

Experiment A compares the double integrator with a predictor fitted to virtual-input–position data. On independent test trajectories, both predict future positions from the same prescribed inputs and past measured positions. Future positions enter only the error calculation. The comparison covers exact and inaccurate mechanics at three noise levels.

Learning gives a modest improvement with exact parameters and a much larger improvement with parameter errors, indicating that data capture systematic departures from the ideal dynamics. The learned predictor remains more accurate at each noise level, although its error grows with noise. Noise also changes the feedback-controlled collection trajectories, so comparisons across noise levels include this effect.

Discussion and Conclusions

This study examined geometric robust data-driven predictive control for a feedback-linearised planar 2R arm. The results identify a useful role for data when the mechanical model captures the main structure but leaves systematic prediction errors. Learning the resulting virtual-input–position response improves prediction and tracking under parameter errors. With exact mechanical parameters, the known double integrator already provides an effective controller. Performance relative to direct regularised data-enabled predictive control depends on the reference, with different trade-offs between accuracy and torque effort.

The predictive settings also remained effective across the tested noise levels and arm parameters. This transfer used newly collected data and corresponding mechanical-model and observer updates. Higher measurement noise increased tracking error and actuator activity, although every geometric data-driven predictive control trial completed.

Generative AI Use

Generative AI tools assisted with literature review, preparation of TikZ illustrations, simulation-code optimisation, and auditing the mathematical exposition and results.

Prediction for Prescribed Inputs

The future output entries are the forecast. For the known double integrator, its exact behaviour and zero affine mean are used in the same solve. Future positions enter only the error calculation.

Double-Integrator MPC

The positive-definite coefficient matrix is factored once; the reference and observer prior update the right-hand side.

References and Performance Measures

The references have phases and a two-second ramp. Evaluation follows a 1.2-second stabilising hold and lasts for the ramp and two cycles: 14 seconds for the base and shifted references, and 11 seconds for the faster reference. Each condition uses six independent data/noise pairs, with measurement noise paired across controllers.

Frequently Asked Questions

What does feedback linearisation do for the planar 2R arm? It uses a mechanical model to map virtual acceleration commands to joint torques. With exact parameters and states, each joint behaves as a double integrator from virtual input to position.

What does the data-driven predictor learn? It learns the virtual-input–position response from recorded data, including residual dynamics caused by parameter, state-estimation, and actuator errors.

When does learning improve prediction and tracking? The learned predictor improves prediction with exact parameters and shows a much larger improvement when mechanical parameters are inaccurate. Learning the response also improves tracking under parameter errors.

How does measurement noise affect performance? Higher noise increases prediction error, tracking error, and actuator activity. Every geometric data-driven predictive control trial nevertheless completed.