Humanoid push recovery logs (simulation)

This dataset was created using a simulator where an humanoid robot was walking.

Some force with a random duration, amplitude and orientation was applied to its trunk. Sometime it fell, and some other it didn't. The goal is to be able to predict as much (and as soon) as possible the fact that the robot is going to fall.

Here are the columns:

t: the time (seconds)

fallen: is the robot fallen (this is not a ground truth and easy to compute from robot state, we want to predict that the robot will be fallen in a near future)

com_x, com_y, com_z: the position of center of mass of the robot in the trunk frame

trunk_pitch, trunk_roll: the roll and pitch (orientation) of the trunk (in reality, it can be measured using IMU)

left,x, left_y, left_weight, right_x, right_y, right_weight: position and intensity of forces below the feet (in reality, it can be measured using force sensors)

left_ankle_pitch,left_ankle_roll,left_hip_pitch,left_hip_roll,left_hip_yaw,left_knee,right_ankle_pitch,right_ankle_roll,     right_hip_pitch,right_hip_roll,right_hip_yaw,right_knee: angle (radian) of robot's joints

force_orientation, force_magnitude: this is a ground truth (unknown in reality, only in simulator) of the force applied on the robot's trunk

Video (YouTube) of the Dataset creation