probabilistic_robotics

The Probabilistic Robotics environment is a simulation focused on autonomous robot navigation, offering two distinct tasks within a discrete action-space and continuous state-space setup.

In the NavigationTrack task, the robot must navigate along a track while avoiding walls. The state-space consists of 5 range measurements, and the action-space allows the robot to move forward, rotate left, or rotate right. Noise is introduced to both actions and states with additive white Gaussian noise. The robot is rewarded for moving forward and penalized for rotation or crashing into the walls.

The NavigationGoal task introduces a dynamic goal location and obstacles, requiring the robot to reach the goal while avoiding crashes. The state-space includes range measurements as well as the distance and angle to the goal. The reward structure incentivizes reaching the goal while penalizing crashes and rewarding the robot based on its progress toward the goal at each step. Both tasks rely on probabilistic factors, such as noise in actions and states, reflecting real-world uncertainties.

When used with the Proximal Policy Optimization (PPO) algorithm, the environment provides a challenging benchmark for training autonomous navigation agents, where the focus is on learning to navigate while managing probabilistic noise and dynamic obstacles.