Weiran Liao

We present BEAST, a novel action tokenizer that encodes action sequences into compact discrete or continuous tokens using B-spline.

We present PointMapPolicy, a multi-modal imitation learning method that conditions diffusion policies on point maps