1. Proposal number: 8 “Towards Learning to Identify Zippers” by Ritika Sahai 2. Should this proposal be considered for the Best Proposal prize? yes 3. On a scale of 1 to 10, how would you rate the overall organization/clarity of the proposal? 8 4. On a scale of 1 to 10, how would you rate the overall project idea? 9 Then write 2-3 pages of helpful feedback to the proposal's author(s). The following questions should help you organize your feedback: * Overall, is the proposal clear, concise, and well-organized? Most parts of the proposal are very clear, organized and informative. * Does the proposal meet the posted proposal guidelines? Yes, it meets the guidelines. * How does the project idea fit within the framework of Developmental Robotics? The project can definitely fit very well within developmental robotics research. Infants also explore objects through scratching and this study can serve as an experiment to test methods that will afford robots a similar ability to perceive surfaces and textures. Another good thing about this proposal is that the robot’s representation of textures will be grounded in its own experience with the world. * Describe what you like BEST and LEAST about the project idea. What I like best about the proposal is the overall direction which builds upon previous work in tactile object perception from Stoytchev’s lab. The application is very relevant, but at the same time it can used as a model for how robots can generally identify discontinuities in object textures. The only thing that I do not like in this proposal is that it skips over a lot of the details regarding how the zippers (or other texture boundaries) are going to be detected. There is good detail describing the experimental platform and exploration behaviors to be performed by the robot; however, there is little detail describing how the sensory data from the tactile sensor will be analyzed and used to detect things like zippers. * Do you have any concerns about the project? Does it seem doable in the remaining time? I don’t see any major concerns and given the author’s previous experience with tactile perception, the project seems doable. * Does it seem too difficult? Maybe not too difficult, but definitely difficult. * Are there any major details left out? Yes, some. For example, there are no details regarding which methods will be used when analyzing the accelerometer data. Is the visual model going to be based on the same features used in Sukhoy’s work on button exploration? Also, zippers make sounds - would auditory data also be used by the robot? * Does the idea rely upon technologies that are not currently available? No. * Do you have any suggestions for improvement? Do you have any suggestions for related work that should be cited? There are several suggestions I can think of. First, it seems that zipper detection is a specific instance of a surface in which one or more surface patches (which may or may not be of the same type) are connected with a boundary. Does the author plan on using the same exploratory behaviors to characterize other types of objects - for example, an open book has a boundary in the middle of the two pages, which perhaps could also be detected if the robot slides its finger from side of the open book to the other. In my opinion, it would be very good if the method can be shown to also solve other tasks in addition to zipper detection. The same method can probably solve the task of detecting the boundary between a piece of cloth laid out on the ironing board and the ironing board itself. A second suggestion is to integrate the previous work on surface recognition and characterization. Since the robot will explore a number of zippers, chances are that the surface patches from different objects will be of different materials and textures. Thus, the robot can only detect zippers, but also detect surface types and perhaps learn to recognize textures or categorize the different surface patches it has perceived. It may even be possible to use the visual model to learn visual surface categories and then correlate the visual types with the tactile types. Also, if the zippers vary in certain dimensions (e.g., thickness, material, etc.) would it possible to classify them into types? This information might be useful if specific zipper types are more likely to occur with specific types of objects, or if different zipper types need to be manipulated diffrerently.