DESC0025149

Air leakage accounts for greater energy losses than any other component of the building envelope and is responsible for more than 4% of all the energy used in the U.S. In most homes, the most significant air leaks are hidden in the attic and basement. These leaks can be difficult to seal because they are often hidden under the insulation. The current state of the art for air sealing a home including attic is manual and expensive. Highly trained air sealing technicians willing to crawl through confined attic spaces are extremely hard to find and very expensive since the job is tedious and hazardous; there is always an imminent danger of accidentally falling through the drywall. Moreover, attics in the summertime could be as hot as 150° F. As a result, a great number of attics in the U.S. are left un-retrofitted. Attic contractors generally agree that the most challenging part of attic air sealing is the initial cleaning and removal of debris and existing insulation. In this project, we will develop automated robotic solutions for attics and basements to (a) remove and vacuum existing insulation in attics and basements (b) blow new insulation. Our solution considers both the hardware design of the robot as well as the autonomy software needed to enable the robot to traverse harsh attic environments full of joists, clutter and debris. In previous work we demonstrated the ability of a tracked robot with two flippers climbing over joists and vacuuming loose insulation via two sets of remote controls both operated by a human user: one to maneuver the base robot over joists and the other to maneuver the vacuum hose. We found this solution to be extremely taxing on the human operator and to result in sensory overload and operator fatigue. In Phase 1 of this project, we develop Rumba like autonomy algorithms to enable the robot maneuver and move in the harsh attic environment full of joists and other clutter. In Phase 2, we design a payload for vacuuming and insulation blowing and develop associated autonomy algorithms for maneuvering the hose. Our autonomy algorithms use a deep learning AI method called reinforcement learning and will leverage our existing solution which uses this method for teaching a hexapod robot to climb joists. The basic approach is to first teach the robot the tasks in simulation and then transfer the learned skill to ôrealityö commonly referred to as sim-to-real. The main benefit to the public is lower cost of attic retrofit. In addition, our solution can service attics that were previously not serviceable by humans due to their tight space. Overall, this technology will allow American homeowners save on energy bills in very cold and very hot climates. We intend to make the equipment available to attic contractors through chain stores such as Home Depot or Lowes both for sales and for renting.