Model Compression for Resource-Constrained Mobile Robots

The number of mobile robots with constrained computing resources that need to execute complex machine learning models has been increasing during the past decade. Commonly, these rely on edge infrastructure accessible over wireless communication heavy computational tasks. However, might become unavailable and, consequently, oblige execution tasks robot. This work focuses making it possible by reducing complexity and total parameters pre-trained computer vision models. is achieved using model compression techniques such as Pruning Knowledge Distillation. These have strong theoretical practical foundations, but their combined usage not widely explored in literature. Therefore, this especially investigating effects combining two techniques. results reveal up 90% a can be removed without any considerable reduction model's accuracy.