- Mode
- Double Ackermann
- Source
- none
- Phase
- HOLD
- Turn
- R → ∞
Arrivals 0
Best 0
- First arrival The rover reached a pose you set.
- All three modes Rolled in double Ackermann, crab and point turn.
- Ten arrivals Reached 10 poses in one visit.
A simplified model of the Gazebo simulation: flat ground, kinematics only, running at 3 times real speed. The planner and steering math are ported from the ROS 2 nodes with the project's own numbers: cruise 0.4 m/s, lookahead 1.2 m, minimum turning radius 0.99 m, and wheels that stay still until every knuckle is within 5 degrees of its commanded angle.
VERTEX rover simulation
Built the ROS 2 and Gazebo simulation of the VERTEX rover and a leveling controller that holds its deck within 1 degree of gravity while driving and turning on a 10-degree slope.
1. Deck normal
The blue arrow points straight out of the deck. Leveling means keeping this arrow lined up against gravity while the ground tilts underneath.
One frame of the recording. Select a number to read what that overlay reports.
The problem
The UMD Space Systems Laboratory is developing BioBot, a rover that carries an astronaut's life-support system and connects to them through an umbilical arm. VERTEX is its Earth-analog chassis. It has to climb 30-degree slopes and cross 40-degree side slopes while keeping its deck level, which means coordinating four steering actuators, four drive motors, and four articulating suspension arms. The engineering purpose is to find control failures in simulation before they reach hardware. My contribution is the simulation and the control software.
What I did
I turned the CAD export into a working ROS 2 and Gazebo model: decimated meshes, inertia derived from geometry, unlocked hip and steering joints, and the control nodes that drive them. I built three steering modes (double Ackermann, crab, and point turn) with a steer-then-drive gate, and a terrain generator that saves a ground-truth height file with every world. I defined level as the angle between the deck normal and gravity, then designed a controller that commands the four hips through two warp-free modes, one for roll and one for pitch.
How it works
The first controllers drove the deck to 23.7 degrees of tilt on mountain terrain. I disconnected the controller, jogged the joints by hand, and traced the divergence to two independent sign errors: mirrored hip axes in the CAD, and a pitch convention in the simulator opposite to my math. A second failure, a stalled steering knuckle and a 0.21 m sensor error, led to an indexing bug in my terrain generator that had turned 99.6% of the mesh into vertical fins. Each fix came from checking an independent measurement against known truth, and the generator now refuses any mesh whose triangles do not tile the ground.
What happened
On a 10-degree slope the rover holds its deck within 1 degree of gravity at rest, driving, and turning. On the fractal mountain it stays within 1 degree for 58% of a lap and recovers within about a second of stopping. Front range sensors match ground truth within 1.7 mm RMS. I present this work at the YPSE conference in November 2026; the next steps are feed-forward leveling from the height map and running the same stack on the physical chassis.