We present Galactic, a large-scale simulation and reinforcement-learning (RL)
framework for robotic mobile manipulation in indoor environments. Specifically,
a Fetch robot (equipped with a mobile base, 7DoF arm, RGBD camera, egomotion,
and onboard sensing) is spawned in a home environment and asked to rearrange
objects - by navigating to an object, picking it up, navigating to a target
location, and then placing the object at the target location.
Galactic is fast. In terms of simulation speed (rendering + physics),
Galactic achieves over 421,000 steps-per-second (SPS) on an 8-GPU node, which
is 54x faster than Habitat 2.0 (7699 SPS). More importantly, Galactic was
designed to optimize the entire rendering + physics + RL interplay since any
bottleneck in the interplay slows down training. In terms of simulation+RL
speed (rendering + physics + inference + learning), Galactic achieves over
108,000 SPS, which 88x faster than Habitat 2.0 (1243 SPS).
These massive speed-ups not only drastically cut the wall-clock training time
of existing experiments, but also unlock an unprecedented scale of new
experiments. First, Galactic can train a mobile pick skill to >80% accuracy in
under 16 minutes, a 100x speedup compared to the over 24 hours it takes to
train the same skill in Habitat 2.0. Second, we use Galactic to perform the
largest-scale experiment to date for rearrangement using 5B steps of experience
in 46 hours, which is equivalent to 20 years of robot experience. This scaling
results in a single neural network composed of task-agnostic components
achieving 85% success in GeometricGoal rearrangement, compared to 0% success
reported in Habitat 2.0 for the same approach. The code is available at
github.com/facebookresearch/galactic.

Galactic 是一个大规模的机器人移动操作仿真和强化学习框架，在家庭环境中通过仿真实现机器人物品重排，获得了显著的性能提升并解锁了前所未有的大规模实验，此外实现了一个具有通用性的神经网络以达到 85% 的抢单重拍成功率。

Galactic：针对每秒 100k 步的重排列的端到端强化学习扩展

Galactic: Scaling End-to-End Reinforcement Learning for Rearrangement at  100k Steps-Per-Second

Recent advances in robotic mobile manipulation have spurred the expansion of
the operating environment for robots from constrained workspaces to
large-scale, human environments. In order to effectively complete tasks in
these spaces, robots must be able to perceive, reason, and execute over a
diversity of affordances, well beyond simple pick-and-place. We posit the
notion of semantic frames provides a compelling representation for robot
actions that is amenable to action-focused perception, task-level reasoning,
action-level execution, and integration with language. Semantic frames, a
product of the linguistics community, define the necessary elements, pre- and
post- conditions, and a set of sequential robot actions necessary to
successfully execute an action evoked by a verb phrase. In this work, we extend
the semantic frame representation for robot manipulation actions and introduce
the problem of Semantic Frame Execution And Localization for Perceiving
Afforded Robot Actions (SEAL) as a graphical model. For the SEAL problem, we
describe our nonparametric Semantic Frame Mapping (SeFM) algorithm for
maintaining belief over a finite set of semantic frames as the locations of
actions afforded to the robot. We show that language models such as GPT-3 are
insufficient to address generalized task execution covered by the SEAL
formulation and SeFM provides robots with efficient search strategies and long
term memory needed when operating in building-scale environments.

本文介绍了一种针对机器人操纵动作的语义框架表示方法，并引入了 SEAL 问题和 SeFM 算法来提高机器人在大型环境中的任务执行效率。