I just thought there needed to be more details in the paper overall.  Some things were not very clear to me immediately, and I don't understand in what applications these sensors would be useful.  

Section IIIa- It would be nice to have a label of roll angle in Fig. 4, like there is for tilt angle.  From the results, I'm not really convinced that you can determine tilt angle that accurately.  You need to know the precise roll angle beforehand, and even if you can assume that roll angle is pre-determined, I don't agree that you can" determine tilt angle before 0,4 radians before becoming ambiguous".  Because some of the plots are not monotonic, I don't think you can truly estimate the tilt angle.  It's impossible, just from these plots, to know whether the measured force corresponds to a tilt angle before or after the peak of the calibration curve.

Section IIIb- the surface exploration test.  I just don't think the plots that they show are clear.  A minor comment is that in their Fig 7 caption, they say the red line marks the surface normal, but there isn't a red line.  It would be nice if they could superimpose what the actual surface looked like onto the plots, so we could see exactly what it was supposed to be detecting.  I just don't think that the plots look much like the bowl they were using, but that could just be the angles at which the plots are being viewed from.  

Section IIIc.  Minor comment:  Axes labels on Fig 9 are awfully small.  Also, they refer to the measurements as "weights", but present the data in grams...if I remember correctly, weight should include gravity and units in grams are more commonly used to measure mass.  Again, looking at the results of this experiment, I'm not convinced that the sensor can be used to estimate the weight, there's a lot of variation and because they only tested one point more than once, I don't know how accurate that curve really is.  Also, doesn't the curve change when a different object with different stiffness is used?  Even if the sensor is only being used to get a rough estimate of weight/force, it's still not that great.  The range of possible weights from the force measurement is pretty large. 

Section IIId. Again, I think that it's hard to conclude you can estimate stiffness.  The stiffness from these measurements would vary from 2500 to 3333 (top of cup) or from 5000 to 10,000 N/m (bottom of cup).  I wonder if it really matters if the stiffness is 5000 or 10,000 N/m, since it's so high to begin with...I guess their main point was that there were 2 different stiffnesses, and the force sensor was able to spout out two separate "lines" of data, but what was the actual stiffness of the cup?  How big was the actual stiffness difference?  

Overall, I remain unconvinced that using these FSR's for useful sensing is possible.  I feel like a little more detail in necessary in the paper to describe their results, and why they feel the sensors have potential.  In what type of application would you want to have extremely rough estimates of forces, but not precise measurements?  A little more discussion of their results might convince me more, but it seemed like they mostly said, here is our experiment, here is our data, it is obvious from the plots that the sensors worked...but for me, it wasn't obvious from many of their plots.