资源描述
2023 11 19 分析师：傅鸿浩 SAC 编号：S1050521120004 分析师：臧天律 SAC 编号：S1050522120001 2 Saycan RT-X 2022 4 Say-can RT-1 SayCan RT-2 VLM RT-1 Open X RT-X 1 2 3D 3+AI YWEVwPmRnOpOsMmPqNoRqR6McM7NmOqQtRoNiNrQnMjMpNoOaQnMsONZsQnPvPqQpM3 目 录CONTENTS2.Saycan RT-X 3.1.4 01 6 ChatGPT LLM corner case ChatGPT for Robotics:Design Principles and Model Abilities 1 7/Reinforcement Learning(Chain of Thought CoT)7 RT-2 VLM 2 RT-2 RT-2:Vision-Language-Action Models 02 9SayCan LLM 2022 4 Say-can Can Do As I Can,Not As I Say 3 LLM Saycan 4 Saycan LLM Do As I Can,Not As I Say 10RT-1 RT-1 image net token,image token Transformer Saycan RT-1 700 13 17 13 internet-scale RT-1:Robotics Transformer for Real-World Control at Scale 5 RT-1 11PaLM-E VLM PaLM 220 ViT-22B PaLM PaLM-E 5620 PaLM-E:An Embodied Multimodal Language Model 6 PaLM-E 7 PaLM-E 562B PaLM-E:An Embodied Multimodal Language Model 12RT-2-VLA RT-2 RT-1 Transformer VLM RT-1/co-fine-tuning VLM RT-2 RT-1/VC-1 RT-2 RT-2:Vision-Language-Action Models 8 RT-2 RT-113RT-2-VLA chain of thought RT-1 SayCan PaLM-E SayCan RT-1 RT-2 RT-2 RT-1 VLM VLA VLM PaLM-E PaLI-X 9 RT-2 ViT LLM RT-2:Vision-Language-Action Models 14RT-X Open X RT-2 RT-1 X-robot Open X 21 22 527 160266 RT-1 RT-2 RT-1-X RT-2-X 10 RT-1 RT-2 RT-1-X RT-2-X Open X-Embodiment:Robotic Learning Datasets and RT-X Models 15RT-X Open X RT-2 RT-1 RT-1-X RT-1 50%RT-1-X RT-1 Open X-Embodiment:Robotic Learning Datasets and RT-X Models RT-2-X RT-2 RT-2-X RT-2 12 RT-2-X 11 RT-1-X RT-116RT-X Open X RT-2 RT-1 RT-X RT-1-X RT-2-X ImageNet Open X-Embodiment:Robotic Learning Datasets and RT-X Models 13 Open X PAGE 17PaLM-E VLM PaLM(LLM)Vision TransformerRobot Transform 1 Robot Transform-2 VLA Robot Transform-X VLA Open X-Embodiment SayCan 14 Open X RT-1 Open X 03 19 Open X VoxPoser IPHONE4 单手臂,48远程操控,6双手机器人,2单手臂远程操控四足机器人滚轮机器人双手机器人 Open X-Embodiment:Robotic Learning Datasets and RT-X Models 15 Open X VoxPoser:Composable 3D Value Mapsfor Robotic Manipulation with Language Models 16 VoxPoser 20 2022 4 SayCan RT-2 action tokenization,VoxPoser value map,SayCan value function,tokenization Model-based RT-2:Vision-Language-Action Models 17 RT-2 token 21/AI-SayCan RT-1 SayCan RT-2 VLM RT-1 SayCan AI A1 WorkGPT LLM VLM AI model-base 3D+RT-2/AI/1 2 3 Robotics modelling,planning and control 18 19 EI Brian 22+1 2 3D RT-1 13 17 3+AI 23 24 6 3 CFA FRM 4 25/12 A 300 1 20%2 10%20%3-10%10%4 10%2-10%10%3-10%报告编号:231119183554
展开阅读全文