Offline Constrained Reinforcement Learning on BulletGym benchmark leaderboard · SOTA2 Research

Offline Constrained Reinforcement Learning on BulletGym

0.76BallRun RewardBCQ-Lag

Evaluation Results

Method	Links
BCQ-Lag2025.05		0.76	3.91	0.94	0.15	0.72	5.54	0.76	5.11	0.69	2.36	0.63	1.89	0.8	3.07	0.58	2.87	0.74	3.11
BC-All2025.05		0.6	5.08	0.97	0.33	0.24	2.13	0.72	2.93	0.74	4.71	0.58	3.74	0.72	3.03	0.58	4.9	0.64	3.36
COptiDICE2025.05		0.59	3.52	0.87	0	0.67	4.15	0.61	0.94	0.7	2.61	0.49	3.14	0.26	1.02	0.17	5.04	0.55	2.55
PDOCRL2025.05		0.48	0.86	0.8	0.11	0.1	0.84	0.03	0	0.61	0.21	0.63	0.11	0.26	0.26	0	0	0.3	0.3
BEAR-Lag2025.05		0.47	5.03	0.68	7.78	0.42	2.47	0.15	0.73	0.86	3.09	0.74	2.18	0.78	3.68	0.65	5.48	0.48	3.8
CDT2025.05		0.39	1.16	0.99	0.65	0.63	0.79	0.72	0.91	0.77	1.07	0.75	0.95	0.63	0.98	0.54	1.78	0.68	1.04
BC-Safe2025.05		0.27	1.46	0.94	0.22	0.28	0.74	0.65	1.09	0.52	0.65	0.5	0.84	0.56	0.57	0.4	0.96	0.52	0.82
CPQ2025.05		0.22	1.27	0.95	1.79	0.33	3.52	0.03	0.02	0.64	0.76	0.71	0.33	0.22	1.28	0	0	0.33	1.12