Some possible directions for future work.
The optimization interface is "user friendly" if the user can make the "best" decisions with the most "relevant" information. (Other definitions are welcome.) Towards making a user friendly interface, one great test case involves decision amongst trajectories to choose from.
- Displays multiple trajectories and their advantages/disadvangaes
- Select trajectory
- Click and drag trajectory
For example, we can visualize state changes over time, or some experiments in
- steepest bank angle display
- fastest point
- does it reach target, etc.
We should provide a suite of agent dynamics.
- double integrator
- linear system
- 6-dof vehicle dynamics (i.e. rocket)
- orbital dynamics
- robotic arm (lie groups)
Other test cases for the "reinforcement learning" crowd.
- cart pole
- contact dynamics (legged)
- etc...
Python is used heavily in robotics, in particular in image recognition or planning. A wrapper for our C/C++ libraries is essential for adoption.
- PY: A library for interfacing with the
socp
andcprs
packages. Possibly throughcvxpy
's standarized interface. - PY: A library for interfacing with the
algorithms
package - MATLAB: interfaces with algorithms package