Data icra2020_imitation_data about ncps

[mlech26l]

HI @huyenng308

The dataset in the file https://github.com/mlech26l/icra_lds/raw/master/icra2020_imitation_data_packed.npz is from a different paper (this here to be precise) and was collected by manually controlling the mobile robot to navigate it through a course of obstacles.

The dataset dimension (678, 32, 541, 1) are:

• 678 samples (=trajectories)
• 32 is the sequence lengths
• 541 is the number of casted LiDAR rays (= number of input features)
• 1 each LiDAR ray returns a single value measuring the distance of the robot to the nearest object in this direction (each LiDAR ray is casted in a different direction)

The labels (678, 32, 1) are

• 678 samples (=trajectories)
• 32 is the sequence lengths
• The direction in which the robot should navigate (e.g., a continuous value representing left to right steering)

The dataset in closed issue #5 is from the NCP paper and concerns the steering of a car and contains both manually controlled trajectories (for training) and trajectories from the algorithm (for testing).

[monika-ntx]

Thank you for your very detailed explanation.

Could you please tell me where to download dataset (car steering- which is generated from the algorithm if I was not mistaken) from repository and please share me the large dataset (193GB)? I would like to access both of them. My email is [email protected]

Thank you so much for your excellent work. I am working on Liquid Time-Constant Networks and it would be really interesting when applying to the task of steering.

[mlech26l]

Hi @huyenng308
I have sent you an email.