Legged millirobots have a wide range of real world applications due to their small size, high mobility, and low manufacturing costs. They are able to go through narrow passages and tunnels which are inaccessible to the larger robots, traverse in complex environments, and have higher potential to climb over obstacles of their sizes compared to wheeled robots. This technical report presents the simulation results of two VelociRoACH hexapedal legged robots collaboratively climbing over a step which cannot be accomplished by one such robot. Two magnets and one tether are added to form and release the connections between two robots in order to help with the robot alignment and provide a tether assist force. When the robots are in each other’s vicinity, magnetic connections will be formed. After the front robot successfully climbs over the obstacle, all connections will be released so that the front robot can continue moving forward. Multiple reward functions are provided during different stages of the training process in order to learn the primitives for step climbing task such as going forward, keeping aligned with each other, pushing and climbing. The experiments demonstrate that after training with different reward functions in the desirable order, the robots are able to climb up the steps with small height variations using the same learned policy. Since a robot can overcome an obstacle taller than itself with the help of other robots, given enough number of such legged millirobots, potentially they are able to climb over any random obstacle by first climbing on top of others. The experiment videos can be found at https://drive.google.com/drive/folders/1XRE5br1szYWwkqelqulygDXDQNhJAILl?usp=sharing. The code can be found at https://github.com/philip324/roach_step_climbing and https://github.com/philip324/rllab-philip.




Download Full History