Intelligent Agent-Based Stimulation for Testing Robotic Software in Human-Robot Interactions

The challenges of robotic software testing extend beyond generic software testing. Realistic, valid, and interesting tests need to be generated for multiple programs and hardware running concurrently, deployed into dynamic environments with people. We investigate the use of Belief-Desire-Intention (BDI) agents as models for test generation, in the domain of human-robot interactions (HRI) in simulations. These models provide human-like rational agency, causality, and a reasoning mechanism. Additionally, we explore suitable exploration methods for BDI models, and their effectiveness in the generation of diverse tests that cover the robotic code, and allow determining if the software meets its functional requirements. In particular, we applied constrained pseudorandom and brute force exploration coupled with manual input to generate more tests, and fully automated exploration through machine learning (reinforcement learning or RL) and coverage feedback loops. We illustrate our novel approach through an example, where tests are generated to stimulate the robotic software indirectly, via stimulating realistic models of the environment the software interacts with. We show that BDI agents provide intuitive and human-like models for test generation in robotic domains such as HRI. We also demonstrate that BDI models and RL are helpful to explore the software under test and to find requirement violations.