Illinois team wins NASA's LEAPFROG challenge
A team from the University of Illinois Urbana-Champaign won the 2021 Lunar Entry and Approach Platform for Research on Ground competition. For the LEAPFROG challenge, funded by NASA’s Artemis Student Challenges Program, Professor Melkior Ornik’s graduate students Hamza El-Kebir served as the team’s student lead and Pranay Thangeda was the navigation lead.
The challenge was to perform lunar lander operations, focused on navigation and control, in a virtual environment that simulates lander behavior and obstacles on the lunar surface.
“We were tasked with controlling a lunar lander, about the size of a mini-fridge,” said El-Kebir. “It had to take off from the lunar surface, lock onto the nearest crater, navigate to that crater, and safely land. The challenge was quite open ended as to how to do this, with teams being given access to only a small number of virtual sensors installed on the vehicle—an accelerometer, a gyroscope, and a laser ranging sensor. We had no knowledge of the terrain beforehand. Controlling the vehicle was done using a number of actuators: a cold-gas thruster-based attitude control system with four thrusters, and the main thruster, with thrust vector control,” he said.
All of the maneuvers had to be done with speed and finesse.
For example, flying above a 20-meter ceiling from the ground received a per-second penalty. Using fuel? --another per-second penalty. If the lander crashes, the flight is disqualified, but landing it close to the center of the crater, gets a big bonus.
“As is common practice in control engineering, we first set out to fully understand what we were working with,” Thangeda said. “We developed a dynamical model of the lander that reflects the actual behavior of the lander in flight. We derived the equations of motion of the vehicle, ending up with a whopping number of 13 control inputs—12 directions of control for the cold-gas thrusters, and one main thruster.”
Early in the challenge the team decided to use only the attitude control system to control the attitude, because from experience they knew that actively changing the direction of the main thruster can come with many complications. They retrieved the vehicle's parameters from the digital vehicle model that specifies mass and inertial properties, as well as the exact actuator placements.
“What we were left with was quite a sizable vehicle model, much like one you would see used for an aircraft,” Thangeda said.
Ornik, who was the team’s faculty adviser, said many teams try to approach the problem as you would if you were controlling a quadcopter, that is, decouple the longitudinal and lateral motion of the vehicle, and control each one separately.
“Just by looking at the model, the team knew this would almost certainly not work; the vehicle was asymmetric, with the main thruster being about two inches off-center,” Ornik said. “This forced them to stick with the full dynamical model, which we suspected most teams did not even consider because of the complexity that comes with it.”
Not knowing the terrain posed another problem—avoiding a crash.
“We came up with a number of ingenious approaches to create a terrain model, but all of them were unrealistic,” El-Kebir said. “They would require too much time, which would translate in large score penalties. We decided to keep things simple and focused on gaining as much altitude as allowed in the first moments to steer clear from any obstacles.”
The flight regime resulted in ascent, cruising/coasting, deceleration, and descent.
El-Kebir said everything looked good on paper, but they still needed a way to get the vehicle to track their flight path.
“Because our model of the flight vehicle was quite nonlinear and complex, it wasn’t possible to obtain one controller to do the trick,” El-Kebir said. “Changing the angle of the vehicle changes what each control input does. This required us to formulate a new control law from scratch at a high rate. We ended up using what is known as a linear quadratic regulator, which is a type of controller that forms the backbone of modern flight control systems. Given a desired point, the LQR controller tells us what control input to apply. After a couple of hours of tuning the model, we came up with a working prototype that we ended up submitting.”
According to Ornik, their biggest challenge was the asymmetry of the vehicle model. “I think most other teams struggled with that, because the only way to keep the vehicle in level flight is by spinning it rapidly. This was something their controller managed to do on its own, because of its high update rate and the effort they put into tuning it.
“We were able to do it from a programming standpoint because of an open-source control library that I have been developing called Lodestar,” El-Kebir said. “Computing a control law in real time is unheard of these days, but because of the efficient routines I put into the software we were able to pull it off. I think this is the main reason why our results turned out the way they did.”
Both Thangeda and El-Kebir agreed that the competition was interesting, partly because they were given little to no information and constraints.
“We had to rely on our own knowledge of control theory and aerospace dynamical systems to fine tune every part of the controller,” Thangeda said.
El-Kebir added, “It also turned out to be a nice proving ground for the Lodestar control library, which performed beyond our expectations, and made programming very enjoyable. We didn’t have to focus on low-level details. The main experience we gained was developing a control law for a complex dynamical system from scratch within a limited time frame and under stringent constraints.”