search

SoMa-Project / ec_grasp_planner

Grasp Planner Based on Environmental Constraint Exploitation
BSD 3-Clause "New" or "Revised" License
7 stars 3 forks source link

Common heuristic interface #48

Open hmn-ocado opened 5 years ago

hmn-ocado commented 5 years ago
  1. Changed _process_objectsecs method to be compatible with both heuristic.
  2. Changed the structure of success and angle in _objectparam.yaml, now we can store success and angle values for different hands. (For the cucumber there are 4 success values and 3 angle values, is it an error?)
  3. The No Grasp Found case (Q_matrix.max() == 0) is detected earlier and returns an empty hybrid automaton.
JannisW commented 5 years ago

Regarding you second point: No that is not an error. The semantic is the following:

angle describes relative object orientation to EC. The "optimal" orientation values (+/- the epsilon parameter) with their respective probabilities are given in the object_param.yaml file. If the actual angle does not fall into one of these optimal intervals. The last value of the list of success values is selected. Or to put it differently, the last element in the success list represents the success probability of any other orientation that does not fall into one of those intervals defined by angle and epsilon.

The case of the punnet uses this explicitly since the last element of the success list is the same for the last angle and the default case. Both are zero. Thats why the number of angles and success probabilities are the same. I see that this might be confusing, so I'll make it explicit in the merged version and also add a description to the object_param.yaml file.

hmn-ocado commented 5 years ago

Ok great, thanks for the clarification!

JannisW commented 5 years ago

Changes from this branch were merged into master, by merging #57