RegressionPolicy-Pi0.5

HT checkpoint for the four LIBERO suites: libero/checkpoint-30000/pretrained_model. Includes the original model, configs, processors, normalization statistics, and tokenizer. Use the matching HT-capable LeRobot runtime. Saved ht_df=2 corresponds to joint nu=700 over 50x7 valid action coordinates.

The weights and tokenizer are subject to the upstream Gemma Terms of Use, including Section 3.2 and the Gemma Prohibited Use Policy. These requirements come from Google's Gemma/PaliGemma components, not additional restrictions imposed by the RegressionPolicy authors. The weights are not licensed under Apache-2.0. See LICENSE, NOTICE, and GEMMA_PROHIBITED_USE_POLICY.txt.