CEAR

All sequences:

• All Indoor Sequences <— Click to download
• All Outdoor Sequences <— Click to download
• All Backflip Sequences <— Click to download

 

---

Data Structure

Below is the data structure for each sequence:

• Event:
  • event_*.aedat4 (* represents the type of event camera)
• IMU:
  • vectornav.txt
    • timestamp(us) gx(rad/s) gy(rad/s) gz(rad/s) ax(m/s^2) ay(m/s^2) az(m/s^2) magx magy magz qw qx qy qz
• RealSense:
  • realsense_timestamp.txt
    • timestamp(us)_depth_rgb.png timestamp_depth_event.png timestamp_rgb.png …
• LiDAR:
  • lidar.bag
    • Type: sensor_msgs/PointCloud2
    • Topic: /velodyne_points
• MoCap:
  • MoCap.txt or FasterLIO.txt
    • timestamp(s) x y z qx qy qz qw
• Joint:
  • mini_cheetah_joint.txt
    • timestamp(us) FrontRight(abd hip knee) FrontLeft(abd hip knee) HindRight(abd hip knee) HindLeft(abd hip knee) All in rad unit
• Raw and Processed Data:
  • raw_rgb stores unprocessed RGB images from RealSense camera.
  • rgb folder stores processed (smoothed/noise-free) RGB images.
  • raw_depth stores depth images in depth camera’s frame.
  • depth folder stores projected depth images in RGB and event cameras frame.

 

---

Sensors for each sequence

	DAVIS346 Event Camera	DVXplorer Event Camera	LiDAR	RGB-D Camera, IMU, Joint Encoders
Indoor sequences	❌	✔️	✔️	✔️
Outdoor sequences except for X	✔️	❌	✔️	✔️
Outdoor sequences X	❌	✔️	✔️	✔️
Backflip sequences	❌	✔️	❌	✔️

• X represents around-building_day_comb, between-buildings_day_comb, downtown1_day_comb, downtown1_day_trot, downtown2_day_comb, downtown2_day_trot, forest_day_comb, forest_day_trot, grass1_day_comb, grass1_day_trot, grass1_night_comb, grass1_night_trot, parking-lot1_day_comb, parking-lot1_day_trot, parking-lot1_night_comb, parking-lot1_night_trot, residential-area_day_comb, residential-area_day_trot sequences.

 

---

Challenges for each seqeuence

We have also identified and categorized the challenging features in the dataset as follows: Agile locomotion (Ag), Dynamic objects (Dyn), High dynamic range (HDR), Darkness (Dark), Direct sunlight (Sun), Blinking light (Blink), Slippery ground (Slip), Featureless regions (Feat), and Foggy weather (Fog)

• Click to check challenges for each sequence

 

---

Indoor sequence example

Indoor sequences were collected in 13 diverse environments under a range of lighting conditions: well-lit, dark, HDR, and blinking. *_trot means a trot-only gait and *_comb means a combination of trotting, bounding, and pronking gaits.

 

---

Outdoor sequence example

The outdoor sequences were captured in 10 distinctive outdoor environments, with four sequences per environment. Each set of four sequences includes variations of two different sets of quadruped gaits (trot-only and combined gaits) under different periods of the day (daytime and nighttime).

 

---

Backflip sequence example

Backflip sequences were captured in 7 indoor environments and 1 outdoor environment, with two sequences in each environment.