gulucaptain/RoboKata-1110

RoboICL + A+B1 + Foresight Indexing

These Safetensors contain RoboICL trainable adapter weights. See adapter_config.json for the base revision, camera encoding, feature schemas, and indexing settings. Each checkpoint has a SHA-256 receipt. Load the indicated Zero-WAM base and the matching RoboKata integration. The semantic future heads are training-only. The frame key heads, demo attention, and indexing parameters are used at inference.