Rocket Landing Control with Grid Fins and Path-following using MPC

TL;DR

The work addresses fuel-efficient rocket landing by first computing an optimal descent with grid fins via batch optimization, then tracking the resulting trajectory with a TOPED-based MPC that accommodates model mismatch and varying initial conditions. Grid fins provide additional lift/torque, yielding faster stabilization of the attitude angle $\theta$ and about 8% fuel savings compared to a baseline without fins. The TOPED controller uses a demonstration trajectory as a reference and adjusts weights to emphasize tracking of horizontal/vertical positions $x,y$ and inputs $F,\delta$, achieving close adherence under disturbances with significantly reduced computation time (approximately 56.75 seconds for MPC vs about 30 minutes for the batch solution). These results suggest practical benefits for real-time guidance while maintaining near-optimal performance, with planned extensions to 3D scenarios and data-driven demonstrations to further generalize behavior.

Abstract

In this project, we attempt to optimize a landing trajectory of a rocket. The goal is to minimize the total fuel consumption during the landing process using different techniques. Once the optimal and feasible trajectory is generated using batch approach, we attempt to follow the path using a Model Predictive Control (MPC) based algorithm, called Trajectory Optimizing Path following Estimation from Demonstration (TOPED), in order to generalize to similar initial states and models, where we introduce a novel cost function for the MPC to solve. We further show that TOPED can follow a demonstration trajectory well in practice under model mismatch and different initial states.

Figures (2)

• Figure 1: States and input changes in optimal landing trajectory for each model. Top row:$x$ and $y$ path during landing, and angle $\theta$ changes. Middle row: Fuel consumption changes and thrust changes. Bottom row: Grid fin coefficient changes.
• Figure 2: Path following trajectory under model mismatch and different initial state. Initial condition 1 is 15 meters farther horizontally and 100 meters farther vertically, with 55kg less fuel to begin with. Initial condition 2 is 10 meters farther horizontally and 100 meters nearer vertically with 100kg less fuel to begin with. Initial condition 3 is 10 meters nearer horizontally and 100 meters farther vertically, with the same amount of fuel. All three test models are injected with random Gaussian noise $\mathcal{N}$.