Measuring Progress in Robotics: Benchmarking and the ‘Measure-Target Confusion’

While it is often said that in order to qualify as a true science robotics should aspire to reproducible and measurable results that allow benchmarking, I argue that a focus on benchmarking will be a hindrance for progress. Several academic disciplines that have been led into pursuing only reproducible and measurable ‘scientific’ results—robotics should be careful not to fall into that trap. Results that can be benchmarked must be specific and context-dependent, but robotics targets whole complex systems independently of a specific context—so working towards progress on the technical measure risks missing that target. It would constitute aiming for the measure rather than the target: what I call ‘measure-target confusion’. The role of benchmarking in robotics shows that the more general problem to measure progress towards more intelligent machines will not be solved by technical benchmarks; we need a balanced approach with technical benchmarks, real-life testing and qualitative judgment.