Share your thoughts, 1 month free Claude Pro on usSee more

Embodied Task Planning on VirtualHome (Seen)

31.2Success Rate

GRPO w/ EVU

Updated 1mo ago

Evaluation Results

Method	Links
GRPO w/ EVU 2026.04		31.2	-	-	-	-	-
PPO w/ EVU 2026.04		28.8	-	-	-	-	-
SFT w/ EVU 2026.04		27.2	-	-	-	-	-
GRPO 2026.04		25.6	-	-	-	-	-
PPO 2026.04		24	-	-	-	-	-
GRPO w/ EVU 2026.04		20	-	-	-	-	-
SFT 2026.04		20	-	-	-	-	-
PPO w/ EVU 2026.04		17.6	-	-	-	-	-
ReAct w/ EVU 2026.04		16	-	-	-	-	-
SFT w/ EVU 2026.04		16	-	-	-	-	-
GRPO 2026.04		15.7	-	-	-	-	-
Plan-and-Act w/ EVU 2026.04		15.2	-	-	-	-	-
ReAct 2026.04		13.6	-	-	-	-	-
NoThinking w/ EVU 2026.04		12.8	-	-	-	-	-
Plan-and-Act 2026.04		12.8	-	-	-	-	-
PPO 2026.04		10.4	-	-	-	-	-
NoThinking 2026.04		8	-	-	-	-	-
SFT 2026.04		7.2	-	-	-	-	-
UCT 2023.05		-	0	0	0	0	0
finetuned GPT2 policy 2023.05		-	8,130	5,900	4,120	3,090	230
GPT3.5 Policy 2023.05		-	8,340	4,700	7,430	4,820	540
GPT3.5-MCTS 2023.05		-	9,140	7,120	8,810	7,260	3,360
ZSP 2026.01		-	10.78	27.81	-	-	-
LLM+FT 2026.01		-	61.37	17.98	-	-	-
LLM-Planner 2026.01		-	50.98	17.85	-	-	-
FLARE 2026.01		-	54.69	19.93	-	-	-
SayCanPay 2026.01		-	64.14	15.7	-	-	-
TMoW 2026.01		-	83.61	11.07	-	-	-