NEO-VLA: Native Vision-Language-Action Policy Without Vision Encoder

LMMs-Lab released NEO-VLA, a native Vision-Language-Action policy that drops the separate pretrained vision encoder used by mainstream VLAs, with effectiveness validated on the RoboTwin benchmark.

2026-11-06 ~ 2026-11-06 · 2 related posts

• NEO-VLA: native VLA policies drop the separate vision encoder, tested on RoboTwin C2R — liuziwei7 · 2026-11-06
• NEO-VLA drops the vision encoder: native VLA pretraining pays off on RoboTwin — liuziwei7 · 2026-11-06