To begin with, it could be used to for bots that use a keyboard and mouse to navigate websites, book flights, or buy groceries online. The researchers claim that their approach could be used to train AI to carry out other tasks. Taking a bot trained with VPT and fine-tuning it with reinforcement learning allowed it to carry out tasks involving more than 20,000 consecutive actions. Even so, the team found that the best results came from using imitation learning and reinforcement learning together. Using VPT, OpenAI’s bot was able to carry out tasks that would have been impossible using reinforcement learning alone, such as crafting planks and turning them into a table, which involves around 970 consecutive actions. MineDojo, a Minecraft environment with dozens of predesigned challenges, won an award at this year’s NeurIPS, one of the biggest AI conferences. Minecraft is becoming an important testbed for new AI techniques. ![]() “We wanted to expand it, and we thought Minecraft was a great domain to work in.” ![]() “The agents kind of took over the universe there was nothing else for them to do,” says Baker. But the bots soon outgrew their surroundings. Baker was one of the researchers behind Hide & Seek, a project in which bots were let loose in a virtual playground where they used reinforcement learning to figure out how to cooperate and use tools to win simple games. Minecraft’s open-endedness makes it a good environment for training AI. ![]() Players are free to do what they like: wandering a computer-generated world, mining different materials, and combining them to make different objects. But Minecraft is a game with no clear goal.
0 Comments
Leave a Reply. |