Robotics Systems Types

Leon Ziegler wrote:

1. Is it desirable to use polar coordinates relative to the robot? In my opinion this only makes sense for data that (more or less) directly represents sensor information. More high-level information like this should be represented in a robot-independent frame. At least as long as there is no coordinate frame transformation system.

The rationale behind having also polar coordinates was that the type is now also usable in situation where you do not have a localization of the robot at all. I'd like to keep this alive.

1. The coordinates distinguish between the robot's head and body as coordinate frame. This implies a certain robot setup and should not be reflected in this type.

That's true. We need a more general structure to represent coordinates in multiple reference frames in parallel. One entry could then be the polar coordinates relative to the robots vision system.

1. The type currently uses 2D coordinates. 3D coordinates would make it more general. The z coordinate could still be set to 0.

Ok

1. I don't like the field name "human_head_direction". A general person hypothesis may have a "orientation", if somebody wants to distinguish between head, gaze and body direction, we should think about a composition of the person hypothesis from different body parts.

This is unfortunately a hard task with the current protobuf syntax. So we currently have not other usable means than containing everything in the data types as optional attributes if we do not want to prevent using such attributes.

1. Is it desirable to have age, gender and identity as "bytes" type? Also, how is "identity" different from tracking_info?

Identity was the trained name of the person. But the presentation of these entries in the data type needs to be refactored.

To sum up, I suggest to use Translation and Rotation types to represent the person's pose in a global way, rename "human_head_direction" and maybe change the age, gender and identity fields.

Additionally I would like to introduce new fields to represent more percepts from which the hypothesis was built. This includes Legs (from laser scan) and skeleton (from kinect tracking). I will submit corresponding protos.

Sebastian suggested to discuss this issue in Monday's Cithack session (2013-03-03, 14:00, central lab).

@Leon, Johannes, Timo: would you be able to attend?

No, HRI workshops

HRI workshops as well

How about next Monday's Cithack session (2013-03-10, 14:00, central lab)?

How does this relate to #1805? Did we address all topics with that issue?