[Paper Review] Performance, Precision, and Payloads: Adaptive Nonlinear MPC for Quadrotors

Drew Hanover, Philipp Foehn+3 more|arXiv (Cornell University)|Sep 9, 2021

Adaptive Control of Nonlinear SystemsEngineering47 references141 citations

TL;DR

This paper proposes L1-NMPC, a hybrid adaptive nonlinear model predictive control framework that enhances quadrotor performance by online learning and compensating for model uncertainties such as unknown payloads, aerodynamic effects, and wind disturbances. By cascading an L1 adaptive controller with a nonlinear MPC, the method reduces tracking error by over 90% compared to non-adaptive NMPC without gain tuning, enabling high-precision flight at speeds up to 70 km/h even under large unknown disturbances.

ABSTRACT

Agile quadrotor flight in challenging environments has the potential to revolutionize shipping, transportation, and search and rescue applications. Nonlinear model predictive control (NMPC) has recently shown promising results for agile quadrotor control, but relies on highly accurate models for maximum performance. Hence, model uncertainties in the form of unmodeled complex aerodynamic effects, varying payloads and parameter mismatch will degrade overall system performance. In this paper, we propose L1-NMPC, a novel hybrid adaptive NMPC to learn model uncertainties online and immediately compensate for them, drastically improving performance over the non-adaptive baseline with minimal computational overhead. Our proposed architecture generalizes to many different environments from which we evaluate wind, unknown payloads, and highly agile flight conditions. The proposed method demonstrates immense flexibility and robustness, with more than 90% tracking error reduction over non-adaptive NMPC under large unknown disturbances and without any gain tuning. In addition, the same controller with identical gains can accurately fly highly agile racing trajectories exhibiting top speeds of 70 km/h, offering tracking performance improvements of around 50% relative to the non-adaptive NMPC baseline.

Motivation & Objective

• To address performance degradation in nonlinear MPC (NMPC) for quadrotors caused by model uncertainties such as unknown payloads, aerodynamic effects, and wind gusts.
• To develop a real-time adaptive control framework that compensates for parametric and non-parametric disturbances without requiring model re-tuning or retraining.
• To enable high-precision, high-speed trajectory tracking under aggressive flight conditions, including 70 km/h racing maneuvers and unknown slung payloads.
• To maintain robustness and flexibility across diverse environments—such as windy conditions or variable mass loads—using a single controller configuration with fixed gains.
• To minimize computational overhead while achieving superior tracking performance compared to state-of-the-art data-driven and model-based MPC methods.

Proposed method

• The method combines a nonlinear model predictive controller (NMPC) with an L1 adaptive controller in a cascaded architecture, where the adaptive layer corrects for model mismatch in real time.
• The adaptation law is derived at the individual rotor thrust level, leveraging the same nonlinear dynamics model used by the NMPC to ensure consistency and stability.
• The adaptive controller computes corrective control signals in under 10 microseconds, enabling fast, real-time compensation with minimal computational overhead.
• The system learns and compensates for unknown disturbances—including unknown payload mass (up to 60% of quadrotor mass) and external aerodynamic forces—without prior knowledge or model updates.
• The controller is transferable across environments: it performs robustly under wind disturbances, unknown slung payloads, and high-agility racing trajectories without re-tuning.
• The approach avoids reliance on accurate aerodynamic models or online retraining of data-driven models, instead using adaptive estimation to correct for unmodeled dynamics.

Experimental results

Research questions

• RQ1 Can an adaptive controller significantly improve tracking performance of NMPC in the presence of unknown model uncertainties such as varying payloads and aerodynamic effects?
• RQ2 To what extent can a single, fixed-gain controller handle diverse disturbances—including wind, unknown masses, and aggressive maneuvers—without re-tuning?
• RQ3 How does the performance of the adaptive NMPC compare to non-adaptive NMPC and state-of-the-art data-driven MPC methods under identical conditions?
• RQ4 Can the adaptive controller effectively compensate for disturbances in real time with minimal computational cost?
• RQ5 Does the method maintain high accuracy during high-speed flight (e.g., 70 km/h) and under extreme accelerations (>4g)?

Key findings

• The L1-NMPC method reduces tracking error by over 90% compared to non-adaptive NMPC when flying with unknown payloads up to 60% of the quadrotor’s mass.
• The controller achieves a 49% relative improvement in tracking performance over the non-adaptive SRT-NMPC baseline on high-speed circular trajectories (up to 36 km/h), even without an aerodynamics model.
• With an unknown slung payload of 13% of the quadrotor’s mass, L1-NMPC achieves 44% higher accuracy than non-adaptive NMPC without a payload.
• The method demonstrates robustness under wind disturbances, maintaining close tracking to reference trajectories even when flying through a fan-generated turbulence field.
• The adaptive controller computes corrective signals in just 10 microseconds, ensuring minimal computational overhead while enabling real-time compensation.
• The same controller with identical gains successfully tracks aggressive racing trajectories at speeds up to 70 km/h, outperforming non-adaptive NMPC by approximately 50% in tracking accuracy.