Embodied Skill Evaluation on BEAR

87.5Task Plan Success (PRG)Human

Evaluation results

Method
Human2025.10	87.5	95.5	92	93.5	0.83	0.77	0.82	92	96.5	94	89	94.5	83.5	88.5	92.5	89.4
GPT-5 w/ BEAR-Agent2025.10	61.67	84.12	75.16	64.05	0.573	0.418	0.447	71.67	84.62	80.81	62.67	71.01	52.17	56.33	42.86	61.29
GPT-5 w/ few-shot in context learning2025.10	61	71.18	64.65	55.88	0.421	0.337	0.363	62	67.95	68.35	49	73.29	51.69	46.33	42.86	54.98
GPT-5 w/ one-shot in context learning2025.10	60.33	70.59	64.01	55.23	0.415	0.33	0.357	61.33	67.31	67.68	49.33	72.64	50.72	46.67	40	54.43
GPT-52025.10	59.67	70	63.69	54.9	0.411	0.326	0.352	61	66.99	67.34	49.67	72.31	50.24	47	40	52.17
GPT-5 w/ Chain-of-Thought prompting2025.10	58.67	67.35	57.19	64.01	0.406	0.321	0.37	61.33	65.38	68.35	52	73.29	47.83	50	34.29	52.78
InternVL3-14B w/ Chain-of-Thought prompting2025.10	44	27.94	21.9	26.92	0.265	0.214	0.213	31.67	44.87	39.39	39.43	41.37	34.69	24.66	0	26.88
InternVL3-14B w/ one-shot in context learning2025.10	42	38.53	28.34	33.33	0.312	0.268	0.287	34	48.72	52.53	29	39.74	31.88	18.67	28.57	33.38
InternVL3-14B w/ BEAR-Agent2025.10	41.67	47.06	31.85	34.97	0.347	0.294	0.269	36.33	53.85	52.86	37	43.65	31.4	26	28.57	36.24
InternVL3-14B2025.10	41	37.94	27.78	32.8	0.304	0.258	0.276	33	51.28	49.49	31.43	43	28.02	21.33	28.57	33.93
InternVL3-14B w/ few-shot in context learning2025.10	41	39.41	28.98	34.31	0.321	0.275	0.297	34	54.49	47.14	34.67	46.25	25.12	24.33	25.71	34.98
Random Choice2025.10	25	—	—	—	—	—	—	25	25	25	25	25	28	25	25	—