Benchmark tasks

Every task drops an autonomous agent into a sandboxed robotics problem and scores the artifact it leaves behind. Filter by platform and research direction; pick a platform to reveal its directions.