Abstract
We study the balancing of a ball with a robotic arm on a platform attached to a Mars rover as a model task. To add realism to the simulation, motion trajectories were artificially generated based on a real terrain grid. These trajectories introduced varying slopes and surface irregularities, allowing the agent to experience more realistic and diverse dynamic conditions during learning. The considered scenario is the basis for applications in tasks such as stabilization of sensitive instruments or liquid samples on planetary exploration vehicles.The robotic arm mounted on the rover, which moves along a given trajectory, effectively stabilizes the ball despite disturbances caused by movement on the terrain. The simulation results confirm that the control policy obtained after training is robust under uncertainty and adaptable to the dynamics of movement.