Suggestions for future work- Purpose of primary integrator is to fuse information coming from two modules. Gesture module provides whole distribution: confidences for all commands present in database. However, Speech SDK used in speech module provides only "Winner" (command with highest confidence) and it's confidence. Therefore, we can only fuse winning commands' confidences from both modules.Possible improvements would be to obtain histogram of confidences for speech as well.
- Include some other commands similar to gothere in Complementary integrator. For more details refer paper [2].
- Problems with synchronization of speech and gesture commands in complementary integrator. Currently in complementary integrator, the speech command has to be done before gesture command. We notice that if gesture command comes first it creates synchronization problem. This issue is yet to be solved. Using mutex instead of sleep(t) could be beneficial.
- Current constraint: in gothere command, angle of rotation in range [-pi/2;pi/2] is not supported. Therefore, user can not point behind him/her. Reason: z coordinate values for arm joints pointing behind the person's torso are unpredictable.
- Currently it is mandatory that only one person is in front of Kinect. Recognition of user should be performed. This would allow robot to obey commands even in crowded environment.
|