松下电器产业株式会社

Vision–Language Models (VLMs) excel at reasoning in linguistic space but struggle with perceptual understanding that requires dense visual perception, e.g., spatial reasoning and geometric awareness. This limitation stems from the fact that current VLMs have limited mechanisms to capture dense visual information across spatial dimensions. We introduce Chain-of-Visual-Thought (CoVT), a framework that enables VLMs to reason not only with discrete text tokens but also through continuous visual tokens—compact latent representations that encode rich perceptual cues. With a small budget of roughly 20 tokens, CoVT distills knowledge from lightweight vision experts that capture complementary properties such as 2D appearance, 3D geometry, spatial layout, and edge structure. During training, the VLM with CoVT autoregressively predicts these visual tokens to reconstruct dense supervision signals (e.g., depth, segmentation, edges, and DINO features). At inference, the model reasons directly in the continuous visual latent space, preserving efficiency while optionally decoding dense predictions for interpretability. Evaluated across more than ten diverse benchmarks, including CV-Bench, MME-RealWorld, MMVP, RealWorldQA, MMStar, WorldMedQA, and HRBench, integrating CoVT into strong VLMs such as Qwen2.5-VL and LLaVA consistently improves performance by 3

The ability to react dynamically to tactile signals has long been considered crucial to agile human-level dexterity. Yet contemporary learning-based Vision-Language-Action (VLA) models for robotic manipulation generally either overlook the tactile modality or are limited to encoders with static cues, due in part to the scarcity of diverse training data and standardized evaluation, architectural constraints in current VLA models, and limitations of static tactile encoders. In this paper, we push the frontier of tactile-reactive manipulation by addressing all of these limitations. We propose a large-scale, 100-hour tactile-rich dataset collected via a novel, data-efficient recipe that prioritizes elementary motor primitives. To effectively exploit naturally high-frequency touch signals without sacrificing the existing capabilities of existing VLAs, we introduce a variable-rate Mixture-of-Transformers (MoT) architecture equipped with a novel temporal tactile VQ-VAE encoder. We demonstrate the effectiveness of tactile-reactive policies on 12 manipulation tasks requiring delicate force control and deformable object manipulation, achieving over 30

This study evaluates, for Japan’s Kyushu area in 2030, the system-wide impacts of optimal demand-side control coordinated with hourly CO2 emission factors, using an integrated supply–demand simulation across multiple scenarios and carbon price levels. Hourly CO2 emission factors reflect the hourly generation mix, which is determined by power‑system operations and varies with electricity demand; therefore, these emission factors are related to consumers’ electricity consumption. The analysis focuses on the value of time-varying CO2 intensity as a control signal: because the emission associated with electricity consumption changes by hour, aligning flexible loads with low-intensity periods can materially reduce total emissions. Modeling approach in this study couples a supply-side operations model with a residential operation model that internalizes hourly CO2 emission factors to electricity price. Flexible demand-side control linked to hourly CO2 emission factors shifts electric-vehicle charging and heat-pump water-heater operation to low-CO2 hours, suppresses peaks in both purchased electricity and sold electricity. Compared with demand-side-only measures and supply-side-only measures, combining such demand-side control with supply-side carbon pricing achieves the largest reductions in both net demand-side and supply-side CO2 emissions. Moreover, the control method helps to reduce generation cost increase due to the carbon pricing. These findings imply that complementing supply-side carbon pricing like carbon levy with demand-side incentives indexed to hourly CO2 emission factors can achieve reduction of both CO2 emissions and generation costs, thereby easing social cost burden.

Robotic manipulation policies often struggle to generalize to novel objects, limiting their real-world utility. In contrast, cognitive science suggests that children develop generalizable dexterous manipulation skills by mastering a small set of simple toys and then applying that knowledge to more complex items. Inspired by this, we study if similar generalization capabilities can also be achieved by robots. Our results indicate robots can learn generalizable grasping using randomly assembled objects that are composed from just four shape primitives: spheres, cuboids, cylinders, and rings. We show that training on these "toys" enables robust generalization to real-world objects, yielding strong zero-shot performance. Crucially, we find the key to this generalization is an object-centric visual representation induced by our proposed detection pooling mechanism. Evaluated in both simulation and on physical robots, our model achieves a 67% real-world grasping success rate on the YCB dataset, outperforming state-of-the-art approaches that rely on substantially more in-domain data. We further study how zero-shot generalization performance scales by varying the number and diversity of training toys and the demonstrations per toy. We believe this work offers a promising path to scalable and generalizable learning in robotic manipulation.

Continuous normalizing flows (CNFs) and diffusion models (DMs) generate high-quality data from a noise distribution. However, their sampling process demands multiple iterations to solve an ordinary differential equation (ODE) with high computational complexity. State-of-the-art methods focus on reducing the number of discrete time steps during sampling to improve efficiency. In this work, we explore a complementary direction in which the quality-complexity tradeoff can also be controlled in terms of the neural network length. We achieve this by rewiring the blocks in the transformer-based architecture to solve an inner discretized ODE w.r.t. its depth. Then, we apply a length consistency term during flow matching training, and as a result, the sampling can be performed with an arbitrary number of time steps and transformer blocks. Unlike others, our ODEt(ODEl) approach is solver-agnostic in time dimension and reduces both latency and, importantly, memory usage. CelebA-HQ and ImageNet generation experiments show a latency reduction of up to 2× in the most efficient sampling mode, and FID improvement of up to 2.8 points for high-quality sampling when applied to prior methods. We open-source our code and checkpoints at github.com/gudovskiy/odelt.