A one-stop VLA toolbox from Dexmal: the unified DexData format spans pretraining, fine-tuning (full/LoRA/RL), inference, and evaluation, supporting mainstream models (π0, CogACT, OFT, MemVLA, GR00T N1) plus the in-house dual-expert DM0; DB-pretraining brings consistent gains across five simulation benchmarks, with 62% average success on RoboChallenge Table30 real-robot evaluation.
A compact generalist navigation model from Light Origins: Qwen3-VL-4B backbone + dual-channel pointing + RVQ action tokens. One checkpoint covers instruction following, object navigation and visual tracking, transferring zero-shot across humanoid/quadruped/wheeled/aerial robots. Trained entirely in simulation, Apache-2.0.
starVLA's representation-centric continued pre-training framework: shallow-layer protection, caption co-training, and OFT+PI+GR00T multi-head co-supervision build a Qwen3-VL-4B action backbone reaching 82.6% on LIBERO-Plus and 92.5% on RoboTwin 2.0 with cross-embodiment transfer.