Event-based Lunar OPtical flow Egomotion estimation (ELOPE) dataset

The "ELOPE" dataset is the official dataset of ESA's Kelvins ELOPE challenge, created in collaboration with:

• European Space Agency (ESA)
• Delft University of Technology (TU Delft)
• University of Adelaide (UoA)

The Purpose of ELOPE is to investigate to what degree it is possible to estimate the velocity of a lunar lander during descent from an optical event stream. Additionally, a rangemeter measurement is provided as well as data from a Inertial Measurment Unit (IMU).

Event cameras (sometimes called neuromorphic vision sensors) allow a novel kind of vision, where pixels correspond to intensity changes and are able to fire indepedently from each other (asychronous). The ELOPE dataset contains synthetic event streams that were generated using high resolution images rendered by the Planet and Asteroid Natural Scene Generation Utility (PANGU) developed by the University of Dundee's Space Technology Centre and post-processed with the v2e camera emulator developed by ETH Zurich. 

The dataset consists of 93 landing sequences, each representing a descend towards a region of the Moon at different times, resulting in challenging lighting conditions and a variety of surface features. The lander is moving from a high gate at about 2-3km to a low gate at about 150m (no touch down) and might perform any number of maneuvers in between.

Our sequences are split into two categories: train and test. Each sequence is available as a compressed numpy-file (.npz-file). After loading such a file, for example

sequence = np.load('0000.npz')

you can access the sequence like a dictionary to obtain the following information:

• events: A sequence of (x,y,p,t) entries, encoding events recorded by a simulated event-camera
  • x: event coordinate [0, 199]
  • y: event coordinate [0, 199]
  • p: polarity [True, False]
  • t: timestamp [µs]
• timestamps: A sequence of timestamps related to the landers motion [s]
• traj: The trajectory of lander for each of the timestamps as a state-vector (x,y,z,vx,vy,vz,phi,theta,psi,p,q,r)
  • x,y,z: position in global frame (the attitude z is given negative by convention) [m]
  • vx, vy, vz: velocity vector [m/s]
  • phi, theta, psi: roll, pitch, jaw Euler angles [rad]
  • p, q, r: corresponding angular velocity [rad/s]
• rangemeter: A sequence (t, d) that represents the distance from the lander to the pixel [100,100] at short intervals 
  • t: timestamps [s]
  • d: distance [m]

The sequences in train have full information about the flight path of the lander and can be used as examples to gain insights into the general problem. The sequences in test have information about position (x,y,z) and velocity (vx,vy,vz) masked out using nan.

More details about the competition setup and solution evaluation are on the Kelvins competition platform.