---
license: apache-2.0
task_categories:
- robotics
tags: [LeRobot, ledrone, uav, drone, sim, pybullet, multi-task, goal-conditioned, state-based, imitation-learning]
size_categories:
- 100K<n<1M
configs:
- config_name: default
  data_files: data/*/*.parquet
---

# ledrone_pybullet_multitask_state

A **multi-task, goal-conditioned, state-based** imitation dataset combining the three
`ledrone` tasks — **takeoff + hover + circle** — into one corpus for training a single
policy. State-only (no FPV video); 13-dim `observation.state` + 3-dim relative goal
`observation.environment_state` + 4-dim `action`. Each frame is tagged with its `task`.

The three tasks are unified by the relative goal (`target − pos`): a single
goal-conditioned policy reacts to how the goal moves (static for hover, ramping for
takeoff, orbiting for circle). Built from the DART-regenerated single-task datasets
(execute-noised / label-clean, slew-limited startup).

| | |
|---|---|
| Episodes | 900 (300 × 3 tasks) |
| Frames | 240,000 |
| Tasks | `takeoff`, `hover`, `circle` |
| State | 13-dim (pos, vel, attitude quat, body rates) |
| Format | LeRobot `v3.0` |

Part of the [ledrone](https://github.com/ahive-org/ledrone) project. Apache-2.0.