Sensory Scaffolding Suite (S3)

该数据集包含了一系列基于机器人的10项任务，旨在评估在感官支架问题设置中的Scaffolder、基线方法以及未来可能的方法。这些任务涵盖了物体拾取、跨越障碍、笔旋转和钢琴演奏等多种动作，且每种任务都有不同的特权传感器和目标传感器设置。这10项任务被设计用来探索各种实际传感器设置的多样性。总体任务目标是评估在各种机器人任务中，在感官支架背景下强化学习代理的表现。

This dataset includes 10 robot-based tasks, designed to evaluate Scaffolder, baseline methods, and potential future approaches within the sensory scaffolding problem setting. These tasks cover diverse actions such as object picking, obstacle crossing, pen spinning, and piano playing, with each task adopting distinct privileged sensor and target sensor configurations. The 10 tasks are constructed to explore the diversity of real-world sensor settings. The overall goal of this dataset is to assess the performance of reinforcement learning agents in the context of sensory scaffolding across various robotic tasks.