机械行业月报：Figure发布第二代人形机器人.docxVIP

%27:aJVLAd4U

trained 1

natio Al

CLIP[1];R3M[2];MVP[3];VIP[4];VC-1[5]:

Voltron[6];RPT[7];GR-1[8];SpawnNet[9]

Pretraining§m-ADynamicsLearning§m-A2eelodMA3ControlPolicy§III-BIMSDUChOD

Pretraining

§m-A

Dynamics

Learning§m-A2

eelodM

A3

ControlPolicy

§III-B

IMSDUChOD)DR-B4

Goal-stateInstructlo?§III-BS

End-in-end§m-Cl

TaskPlanner

fm-c

Language-based§III-C2

Code-based§m-C3

Dreamer[l6)iMWM[17];Iso-Dream[18];TWM[19];IRIS[20]:SWIM[21];DECKARD[22]

Non-TFM

§III-BI

CLlPort[23];BC-Z[2d];MCIL[25]:HULC[26];UniPi[27]

LanguageInstruction§III-BI;§III-B2;§III-B3TFM-based§III-B2Languagecosts[28]

LanguageInstruction

§III-BI;§III-B2;§III-B3

TFM-based

§III-B2

PerAct[31];Gatn[32];RT-I[33];Q-Transformer[34)iSUDD[35]

LLM-based

§III-B3

RT-2[36];RT-X[37];RT-H[38]i

RoboFlamingo[39];VoxPoser[40]

VIMA[41];MOO[42];Octo[43]

RoboCat[44];RT-Trajectory[45];Diffusionpolicy[46]

SayCan[47];LID[48];Translated(LM)[49]:(SL)*

[50];EmbodiedGPT[51];PaLM-E[52];LOO[53]

InnerMonologue[54];LLM-Planner[55];SocraticModels[56]

ProgPrompt[57];ChatGPTforRobotics[58];Codeaspolicies[59]:DEPS[60];ConceptGmphs[61]

^ ,:Ma,Y.,Song,Z.,Zhuang,Y.,Hao,j.,King,I.(2024).ASuaveyonVision-Language-ActionModelsforEmbodiedAI.armypreprint

arXiv:2405.14093.,am$i?irgtP

“m ’, ’y ‘a :“ n.? Z mm e