Potential-based multiobjective reinforcement learning approaches to low-impact agents for AI safety

Peter Vamplew, Cameron Foale, Richard Dazeley, Adam Bignold