Assess the accuracy of robot's understanding of a voice command and its interpretation as a task.
| Associated Requirements | Untitled (https://heathered-english-a1d.notion.site/7bf922f46af548eba06b9167e41e23ba) |
|---|---|
| Equipment | Stretch RE1 |
| Elements | |
| Personnel | Lead: <lead> |
| Location | AI Makerspace, Tepper School of Business |
Turn the robot on and home it.
Activate the speech recognition pipeline.
Praveen to play the role of a patient and will speak (in a normal voice) a pre-defined sentence from our test verification script (ground truth). This is a task that can either be pick-and-place, navigating or just a voice reply.
The robot acts on the voice command.
The robot metric data highlights the task being performed. This is noted as the experimental data.
The test verification script compares the task being performed (experimental data) and the desired task in the script that Praveen implied (ground truth). If the two things don't match, a flag is raised.
Steps 3-6 are repeated 15 times.
Once step 7 is finished, kill the script, and turn the robot off.
The robot was successfully able to understand the tasks that the patient (Praveen) delegated. Out of 15 attempts, the flags were less than 6, meaning that the robot was successfully able to interpret 10 speech templates as task.