Search for: All records

We introduce RIPT-VLA, a simple and scalable reinforcement-learning-based interactive post-training paradigm that fine-tunes pretrained Vision-Language-Action (VLA) models using only sparse binary success rewards. Existing VLA training pipelines rely heavily on offline expert demonstration data and supervised imitation, limiting their ability to adapt to new tasks and environments under low-data regimes. RIPT-VLA addresses this by enabling interactive post-training with a stable policy optimization algorithm based on dynamic rollout sampling and leave-one-out advantage estimation. RIPT-VLA has the following characteristics. First, it applies to various VLA models, resulting in an improvement on the lightweight QueST model by 21.2%, and the 7B OpenVLA-OFT model to an unprecedented 97.5% success rate. Second, it is computationally efficient and data-efficient: with only one demonstration, RIPT-VLA enables an unworkable SFT model (4%) to succeed with a 97% success rate within 15 iterations. Furthermore, we demonstrate that the policy learned by RIPT-VLA generalizes across different tasks and scenarios and is robust to the initial state context. These results highlight RIPT-VLA as a practical and effective paradigm for post-training VLA models through minimal supervision. 

Using methane as a reagent to synthesize high-value chemicals and high-energy density fuels through C−C coupling has attracted intense attention in recent decades, as it avoids completely breaking all C−H bonds in CH4. In the present study, we demonstrated that the coupling of HCHO with the CH3 species from CH4 activation to produce ethanol can be accomplished on the single Pd atom−In2O3 catalyst based on the results of density functional theory (DFT) calculations. The results show that the supported single Pd atom stabilizes the CH3 species following the activation of one C−H bond of CH4, while HCHO adsorbs on the neighboring In site. Facile C−C coupling of HCHO with the methyl species is achieved with an activation barrier of 0.56 eV. We further examined the C−C coupling on other single metal atoms, including Ni, Rh, Pt, and Ag, supported on In2O3 by following a similar pathway and found that a balance of the three key steps for ethanol formation, i.e., CH4 activation, C−C coupling, and ethoxy hydrogenation, was achieved on Pd/In2O3. Taking the production of acetaldehyde and ethylene on the Pd/In2O3 catalyst into consideration, the DFT-based microkinetic analysis indicates that ethanol is the dominant product on the Pd/In2O3 catalyst. The facile C−C coupling between HCHO and dissociated CH4 makes formaldehyde a potential C1 source in the conversion and utilization of methane through an energy- and atom-efficient process. 

This study investigates the potential use of circulating extracellular vesicles’ (EVs) DNA and protein content as biomarkers for traumatic brain injury (TBI) in a mouse model. Despite an overall decrease in EVs count during the acute phase, there was an increased presence of exosomes (CD63+ EVs) during acute and an increase in microvesicles derived from microglia/macrophages (CD11b+ EVs) and astrocytes (ACSA-2+ EVs) in post-acute TBI phases, respectively. Notably, mtDNA exhibited an immediate elevation post-injury. Neuronal (NFL) and microglial (Iba1) markers increased in the acute, while the astrocyte marker (GFAP) increased in post-acute TBI phases. Novel protein biomarkers (SAA, Hp, VWF, CFD, CBG) specific to different TBI phases were also identified. Biostatistical modeling and machine learning identified mtDNA and SAA as decisive markers for TBI detection. These findings emphasize the importance of profiling EVs’ content and their dynamic release as an innovative diagnostic approach for TBI in liquid biopsies