My PhD thesis, exploring the UESMANN neural network architecture which can learn two (or possibly more) functions in a single set of weights and can be trained using gradient descent.
A look at how well UESMANN performs in a homeostatic robot problem.
A simple method for balancing motor load in wheeled rovers using a simulation of cytokine mediated allodynia: simply put, when a particular wheel motor "gets sore" (i.e. has a high operating temperature) the rover lowers the power to it.