The robot called "DreamWalker" features reinforced learning, which is a type of learning method that aims to teach AI the best action to take in certain situations. The multi-limed robot is capable of stably and swiftly going up and down stairs and freely moving on an air mattress. The solution developed by the Korea Advanced Institute of Science & Technology (KAIST) was demonstrated on different deformable terrains to learn how to overcome grounds with an unstable foothold.
KAIST said in a statement that DreamWalker won first place at the event called "Quadruped Robot Autonomous Walking Contest (QRC)." At the five-day event that began on May 29 in London, the South Korean institute competed with 10 teams including the Massachusetts Institute of Technology (MIT) and Carnegie Mellon University (CMU). In the final match, DreamWalker completed its mission in 41 minutes and 52 seconds while the average completion time of other teams that used remote control was about 49 minutes.
KAIST said its robot is expected to be used in environments caused by disasters such as fires as it can operate in uneven environments without assistance from visual and tactile sensors. DreamWalker estimates its location and recognizes obstacles by combining information from various devices including cameras, 3D light detection and ranging (LiDAR) sensors, and inertial measurement units.
According to Myung Hyun, the research team's leader specializing in electrical and electronic engineering, all technologies including the function that recognizes the robot's surrounding environment and the one that enables DreamWalker to find proper routes were developed by the KAIST members.