Search papers, labs, and topics across Lattice.
The paper introduces G0.5, a novel autoregressive Vision-Language-Action (VLA) model that integrates reasoning and action generation within a single transformer decoder, overcoming the limitations of traditional architectures that separate these functions. By employing a learnable action tokenizer, a chain-of-thought stream, and a visual memory module, G0.5 effectively maps diverse robot actions and retains contextual history, enabling it to follow instructions and adapt to various tasks without additional training. The model outperforms state-of-the-art benchmarks across multiple real-world and simulated environments, achieving significant improvements in task execution and adaptability.
G0.5 achieves unprecedented performance in robot reasoning and action by merging decision-making and execution into a single autoregressive framework, outperforming existing models across seven challenging benchmarks.
The prevailing recipe for Vision-Language-Action (VLA) models couples a pretrained VLM with a separately trained flow-matching action expert. This makes the VLM a context encoder rather than a decision-maker. We introduce G0.5, a pretrained autoregressive VLA in which a single transformer decoder emits reasoning and action tokens under a single objective. Three components make this tractable at foundation-model scale: a learnable cross-embodiment action tokenizer that maps heterogeneous robot actions into a shared vocabulary; a native chain-of-thought stream interleaving task decomposition, object grounding, and action hints with action tokens; and a visual memory module that injects multi-second history through the vision encoder. Because reasoning and action share a single set of weights, the pretrained VLM's capabilities carry over to physical behavior: the model follows instructions closely, and prompts directly steer action granularity, task horizon, and out-of-distribution scene handling without further training. Pretrained on a large collection of robot datasets together with VQA samples, G0.5 surpasses state-of-the-art models across 7 independent regimes: real-world fine-tuning on R1lite and R1pro robots (76.7\% vs.\ 53.3\% for $\pi_{0.5}$ and 24.4\% for GR00T-N1.7), the 2025 BEHAVIOR Challenge on 50 long-horizon household mobile manipulation tasks using a generalist policy (31.4\% vs.\ 26.3\% for $\pi_{0.5}$ and 26.1\% for the challenge winner), DROID post-training followed by zero-shot transfer to an unseen environment and objects (82.5\%), a language-following Pick-and-Place benchmark, LIBERO (98.9\%), RoboTwin 2.0 (93.3\%), and SimplerEnv-Bridge (87.3\%).
