Impact-Aware MPC Lets UAVs Land on Heaving Platforms

> Paper: Impact-Aware Model Predictive Control for UAV Landing on a Heaving Platform > Authors: Jess Stephenson, Melissa Greeff > arXiv: 2604.21078 | 2026-04-28

1. Landing on a Moving Deck

Imagine a drone trying to land on a naval ship. Waves make the deck rise and fall continuously.

If the drone simply descends as if landing on solid ground:

• A rising deck slams into it → huge impact forces
• A falling deck drops away → the drone misses and crashes into the sea
• Contact-induced bounce → the UAV rebounds and loses control

Landing a UAV at sea is one of the hardest problems in robot control.

2. MPC: Control That Predicts the Future

The core idea of Model Predictive Control (MPC): instead of deciding based on the current state, predict the system's behavior over a future horizon and optimize the whole trajectory.

It's like driving: you don't look only at the road directly ahead, but at the curves, vehicles, and pedestrians dozens of meters ahead, then plan your entire path.

But traditional MPC has a blind spot: it does not model impact.

3. The Impact-Aware Breakthrough

The key innovation of this research: explicitly modeling the impact process within the MPC framework.

Specifically:

1. Landing is modeled as a rigid-body impact at the velocity level 2. Newton's restitution law describes the post-contact rebound behavior 3. The impact constraints are embedded as a linear complementarity problem (LCP) 4. MPC optimizes the trajectory while simultaneously considering flight dynamics and impact dynamics

This means the UAV "anticipates" the physics of the contact instant when deciding when, where, and at what speed to land.

4. Why This Is Harder Than a "Soft Landing"

"Soft landing" usually means minimizing contact speed. But on a heaving deck:

• Descending too slowly: the deck may sink away before contact, causing a miss
• Descending too fast: the impact may damage the drone or the deck
• The optimal strategy may be to track the deck's motion and make contact at the moment of minimum relative velocity

Impact-aware MPC can find this dynamic optimum—not minimizing absolute speed, but minimizing the relative impact at the moment of contact.

5. A Feynman-Style Judgment: Physical Constraints Cannot Be Bypassed

Feynman emphasized in his physics lectures:

> "Nature cannot be fooled."

In robot control, this means:

• You can optimize trajectories with algorithms
• You can sense the environment with sensors
• But you cannot violate Newton's laws

The success of impact-aware MPC lies in this: it does not try to "bypass" physics, but incorporates physical constraints directly into the optimization problem.

6. Takeaways

If you design robot control systems, ask yourself:

1. "Does my control framework account for all relevant physical processes?" 2. "Have I modeled unavoidable constraints (like impact) as part of the optimization problem?" 3. "Is my prediction horizon long enough to capture key dynamic interactions?" 4. "Have I validated the system's robustness under different environmental conditions?"

Impact-aware MPC teaches us: in the physical world, the best control doesn't avoid all difficulties—it understands and exploits physical laws to handle them gracefully.

When a drone learns to "anticipate" impact, it can land safely even in the most turbulent environments.