The RL framework supports several task environments, each defined by a Gymnasium-compatible class with its own observation space, action space, and reward function. Three pages here cover the main families.
Task: hovering is the simplest — keep the drone in place at a target altitude. It’s the smoke test that the entire pipeline (env → policy → action publisher → simulated drone) is correctly wired up.
Task: obstacle avoidance is the indoor flight task — navigate through a cluttered environment without collisions. This is the task that all 24 dev-log experiments target, refined into the 2D coverage simulator that runs on top of warehouse_v2.
Task: waypoint is a navigation primitive — fly to a specified position and hold. Useful as a building block for hierarchical setups where a top-level planner picks waypoints and a bottom-level policy executes them.
Contents
Auto-generated from child entries during build (update-indexes.mjs).