ant_grasp_matlab

Position Control

Originally this model used the discrete time steps to plot the approach and grasp of a proposed grasp. However as this was for visual information and wasn't required to generate a synthetic grasp, the movement aspect gradually drifted out of the code and the noise/error was introduced manually (using ray intersection) rather than through the URDF and collision calculation. The main difficulty arose from trying to generate a reasonable grasp approximation using the URDF, and collision when the CollisionObject used to estimate antennal collision is essentially an initial approximation before the actual location is generated using triangle-ray intersection. If you really do want this functionality, you basically need to add back in a point where the position path is generated and don't enable grasp_complete in the Ant class, until you want to exit the trial loop.

Mandible Control

This makes the following things redundant (and therefore useful if you want to reintroduce the movement back into the model): Ant:mandible_state PoseControlClass:moveMandible

As of 26/06/23 these functions were tested and work when the mandible_state is varied.

Non-convex shapes

This was always the plan to include this in the model, but the model has been developed in a very iterative sense and generally concave shapes are difficult to generate contacts for. While it is possible to render them in the model, it is by externally separating concave shapes into a set number of convex triangulations and importing that set of convex sub-shapes. The surface normal and point of collision works fairly well for these clusters of shapes but because of how the calculation occurs, the simulation is a lot slower as it checks every sub-shape for collisions individually. There is surely a way that the process could be sped up but it wasn't significant enough for the research at the time to work on it. There are still a few issues with grasping these concave shapes. The latest is that the grasp attempt that is approximated using ray projection does not actually plot (or calculate) the locations of the grasp contacts on the surface of the multi-stl.

Some work has been added to ensure that the grasp evaluations work with multiple STLS, which could do with some more robust testing, however most of the inaccuracies come from poor selection of grasp approach.

obj.RUNTIME_ARGS

This is basically a copy of the construction instructions for the model, and ideally all significant variables are copied from the class so the variable can be changed in the specific class, but also it's not ideal if RUNTIME_ARGS.PLOT is enabled in some classes but not all of them, so some variables are hidden within the local obj.RUNTIME_ARGS copy. If in doubt, check the class initialisation to see what is copied out.

Neck Class

This class is mostly outdated, but the code to add a new neck trajectory based on the goal still works. The neck joints vary and all position and rotation of the full head is accounted for in the PositionControl class. The functions: PoseControlClass:newNeckTrajectory PoseControlClass:moveNeck Neck:findIKforNeckPose Are the relevant ones to analyse for re-introducing the neck rotation functionality.