Oct. 2019 - Jun. 2020
Hierarchical reinforcement learning algorithms typically use a high-level controller to compose sub-policies sequentially or combinationally. Inspired by work on ensembles, we developed an algorithm, ACBC (Actor-Critic By Committee), in which outputs from multiple sub-policies, or experts, are composed with a linear combination. As the experts can control their weights, ACBC performs both sequential and combinational composition without a high-level controller. This work was carried out in UCI's Intelligent Dynamics Lab.
Falls are one of the leading causes of injury for the elderly. To help solve this problem, Safeline uses a Raspberry Pi camera to continuously scan for falls. Once a fall is detected, Safeline alerts connected users by pushing a notification to their iPhones. These users can then seek medical assistance for the fall victim.DevPost GitHub Website