3 ms·
GPT-4 designs reward functions for robot dexterity at Super-human level
- famouswaffles 3y agoParticularly Interesting, "In addition, by examining the average correlation by task (App. E), we observe that the harder the task is, the less correlated the EUREKA rewards. We hypothesize that human rewards are less likely to be near optimal for difficult tasks, leaving more room for EUREKA rewards to be different and better. In a few cases, EUREKA rewards are even negatively correlated with human rewards but perform significantly better, demonstrating that EUREKA can discover novel reward design principles that may run counter to human intuition"