Master of Science in Robotics Thesis Talk

  • Remote Access Enabled - Zoom
  • Virtual Presentation - NEW DATE
  • Masters Student
  • Robotics Institute
  • Carnegie Mellon University
Master's Thesis Presentation

Learning-based modular framework for environment-adaptive planning in exploration tasks

Search-based path planning has spawned a number of different solutions using different paradigms and strategies, both generalized and specific to certain problems, representations, and environments. Split into heuristic and non-heuristic based approaches, where heuristic-based approaches, embedded within these approaches are hyperparameters that control aspects of the planner strategy. For example, in A* the inflation factor hyperparameter determines how greedy the search behaves. Looking only at computation time, while a higher inflation factor is quicker; suboptimal paths compared to a lower inflation factor can be generated. 

We pose the hypothesis that the hyperparameters for best performance are dependent on the type of environment determined by its “clutteredness”. Usually, these hyperparameters are manually tuned by trial and error for a general estimate of the type of environments encountered by the agent and held static through the task. If the map consists of varying environments the absolute best performance of the planner on every map in the dataset is unguaranteed. The direct implementation of such strategies starts afresh on the map under testing; there is no usage of prior experience, except what may be encoded into the hyper-parameters, which is unguaranteed to have the best performance over the varying environment of the exploration task. This problem becomes evident and significant in exploration tasks, where the agent could encounter different environments during the same task. 

For exploration tasks, especially search-and-rescue, efficient navigation is critical, which is to have the best performance in every portion of the map explored while looking for objects of interest. To address this issue this work presents a modular framework to utilize the experience of the underlying structure and the corresponding planning strategies performances of prior environments to predict the unknown map and adapt the planning strategy for maintaining high performance throughout the current exploration task.

Thesis Committee:
Matthew Travers (Co-Advisor)
Howie Choset (Co-Advisor)
Maxim Likhachev
Shohin Mukerjee 

Zoom Participation Enabled. See announcement

For More Information, Please Contact: