Abstract--We present a tactile-sensorized Fin-Ray finger that enables simultaneous detection of contact location and indentation depth through an indirect sensing approach. A hinge mechanism is integrated between the soft Fin-Ray structure and a rigid sensing module, allowing deformation and translation information to be transferred to a bottom crossbeam upon which are an array of marker-tipped pins based on the biomimetic structure of the T acTip vision-based tactile sensor . Deformation patterns captured by an internal camera are processed using a convolutional neural network to infer contact conditions without directly sensing the finger surface. The finger design was optimized by varying pin configurations and hinge orientations, achieving 0.1 mm depth and 2 mm location-sensing accuracies. The perception demonstrated robust generalization to various indenter shapes and sizes, which was applied to a pick-and-place task under uncertain picking positions, where the tactile feedback significantly improved placement accuracy. Overall, this work provides a lightweight, flexible, and scalable tactile sensing solution suitable for soft robotic structures where the sensing needs situating away from the contact interface. I. INTRODUCTION Tactile sensing is essential for achieving dexterous manipulation in robotic hands [1], [2]. For example, to perform delicate tasks like gently grasping and placing eggs or glass plates, humanoid robots such as Figure's F.02 and Tesla's Optimus will need fingertip-mounted tactile sensors to become truly capable [3]. To enhance robotic dexterity, researchers have developed vision-based tactile sensors (VBTSs) that take advantage of recent advancements in computer vision [4]-[7].

Recently, augmenting vision-language-action models (VLAs) with world-models has shown promise in robotic policy learning. However, it remains challenging to jointly predict next-state observations and action sequences because of the inherent difference between the two modalities. To address this, we propose DUal-STream diffusion (DUST), a world-model augmented VLA framework that handles the modality conflict and enhances the performance of VLAs across diverse tasks. Specifically, we propose a multimodal diffusion transformer architecture that explicitly maintains separate modality streams while enabling cross-modal knowledge sharing. In addition, we propose training techniques such as independent noise perturbations for each modality and a decoupled flow matching loss, which enables the model to learn the joint distribution in a bidirectional manner while avoiding the need for a unified latent space. Furthermore, based on the decoupled training framework, we introduce a sampling method where we sample action and vision tokens asynchronously at different rates, which shows improvement through inference-time scaling. Through experiments on simulated benchmarks such as RoboCasa and GR-1, DUST achieves up to 6% gains over a standard VLA baseline and implicit world-modeling methods, with our inference-time scaling approach providing an additional 2-5% gain on success rate. On real-world tasks with the Franka Research 3, DUST outperforms baselines in success rate by 13%, confirming its effectiveness beyond simulation. Lastly, we demonstrate the effectiveness of DUST in large-scale pretraining with action-free videos from BridgeV2, where DUST leads to significant gain when transferred to the RoboCasa benchmark.

A major field of industrial robot applications deals with repetitive tasks that alternate between operating points. For these so-called pick-and-place operations, parallel kinematic manipulators (PKM) are frequently employed. These tasks tend to automatically run for a long period of time and therefore minimizing energy consumption is always of interest. Recent research addresses this topic by the use of elastic elements and particularly series elastic actuators (SEA). This paper explores the possibilities of minimizing energy consumption of SEA actuated PKM performing pick-and-place tasks. The basic idea is to excite eigenmotions that result from the actuator springs and exploit their oscillating characteristics. To this end, a prescribed cyclic pick-and-place operation is analyzed and a dynamic model of SEA driven PKM is derived. Subsequently, an energy minimizing optimal control problem is formulated where operating trajectories as well as SEA stiffnesses are optimized simultaneously. Here, optimizing the actuator stiffness does not account for variable stiffness actuators. It serves as a tool for the design and dimensioning process. The hypothesis on energy reduction is tested on two (parallel) robot applications where redundant actuation is also addressed. The results confirm the validity of this approach.

Increasingly large datasets of robot actions and sensory observations are being collected to train ever-larger neural networks. These datasets are collected based on tasks and while these tasks may be distinct in their descriptions, many involve very similar physical action sequences (e.g., 'pick up an apple' versus'pick up an orange'). As a result, many datasets of robotic tasks are substantially imbalanced in terms of the physical robotic actions they represent. In this work, we propose a simple sampling strategy for policy training that mitigates this imbalance. Our method requires only a few lines of code to integrate into existing code-bases and improves generalization. We evaluate our method in both pre-training small models and fine-tuning large foundational models. Our results show substantial improvements on low-resource tasks compared to prior state-of-the-art methods, without degrading performance on high-resource tasks. This enables more effective use of model capacity for multi-task policies.

Vision-Language-Action models (VLAs) have demonstrated remarkable performance in robot control. However, they remain fundamentally limited in tasks that require high precision due to their single-inference paradigm. While test-time scaling approaches using external verifiers have shown promise, they require additional training and fail to generalize to unseen conditions. We propose Masking Distribution Guided Selection (MG-Select), a novel test-time scaling framework for VLAs that leverages the model's internal properties without requiring additional training or external modules. Our approach utilizes KL divergence from a reference action token distribution as a confidence metric for selecting the optimal action from multiple candidates. We introduce a reference distribution generated by the same VLA but with randomly masked states and language conditions as inputs, ensuring maximum uncertainty while remaining aligned with the target task distribution. Additionally, we propose a joint training strategy that enables the model to learn both conditional and unconditional distributions by applying dropout to state and language conditions, thereby further improving the quality of the reference distribution. Our experiments demonstrate that MG-Select achieves significant performance improvements, including a 28%/35% improvement in real-world in-distribution/out-of-distribution tasks, along with a 168% relative gain on RoboCasa pick-and-place tasks trained with 30 demonstrations.

Quadruped-based mobile manipulation presents significant challenges in robotics due to the diversity of required skills, the extended task horizon, and partial observability. After presenting a multi-stage pick-and-place task as a succinct yet sufficiently rich setup that captures key desiderata for quadruped-based mobile manipulation, we propose an approach that can train a visuo-motor policy entirely in simulation, and achieve nearly 80\% success in the real world. The policy efficiently performs search, approach, grasp, transport, and drop into actions, with emerged behaviors such as re-grasping and task chaining. We conduct an extensive set of real-world experiments with ablation studies highlighting key techniques for efficient training and effective sim-to-real transfer. Additional experiments demonstrate deployment across a variety of indoor and outdoor environments. Demo videos and additional resources are available on the project page: https://horizonrobotics.github.io/gail/SLIM.

Embodied AI Research T eam National Institute of AIST Tokyo, Japan muha.muttaqien@aist.go.jp Embodied AI Research T eam National Institute of AIST Tokyo, Japan tomohiro.motoda@aist.go.jp Embodied AI Research T eam National Institute of AIST Tokyo, Japan ryo.hanai@aist.go.jp Abstract --Robotic pick-and-place tasks in convenience stores pose challenges due to dense object arrangements, occlusions, and variations in object properties such as color, shape, size, and texture. These factors complicate trajectory planning and grasping. This paper introduces a perception-action pipeline leveraging annotation-guided visual prompting, where bounding box annotations identify both pickable objects and placement locations, providing structured spatial guidance. Instead of traditional step-by-step planning, we employ Action Chunking with Transformers (ACT) as an imitation learning algorithm, enabling the robotic arm to predict chunked action sequences from human demonstrations. We evaluate our system based on success rate and visual analysis of grasping behavior, demonstrating improved grasp accuracy and adaptability in retail environments. Robotic pick-and-place tasks are essential in various industrial and retail applications, particularly in convenience stores where robots must handle a diverse range of products with different shapes, sizes, textures, and colors, as shown in Figure 1. However, real-world pick-and-place scenarios pose significant challenges due to dense object arrangements, frequent occlusions, and the need for precise grasping and placement.

Handling objects with unknown or changing masses is a common challenge in robotics, often leading to errors or instability if the control system cannot adapt in real-time. In this paper, we present a novel approach that enables a six-degrees-of-freedom robotic manipulator to reliably follow waypoints while automatically estimating and compensating for unknown payload weight. Our method integrates an admittance control framework with a mass estimator, allowing the robot to dynamically update an excitation force to compensate for the payload mass. This strategy mitigates end-effector sagging and preserves stability when handling objects of unknown weights. We experimentally validated our approach in a challenging pick-and-place task on a shelf with a crossbar, improved accuracy in reaching waypoints and compliant motion compared to a baseline admittance-control scheme. By safely accommodating unknown payloads, our work enhances flexibility in robotic automation and represents a significant step forward in adaptive control for uncertain environments.

Imitation learning for acquiring generalizable policies often requires a large volume of demonstration data, making the process significantly costly. One promising strategy to address this challenge is to leverage the cognitive and decision-making skills of human demonstrators with strong generalization capability, particularly by extracting task-relevant cues from their gaze behavior. However, imitation learning typically involves humans collecting data using demonstration devices that emulate a robot's embodiment and visual condition. This raises the question of how such devices influence gaze behavior. We propose an experimental framework that systematically analyzes demonstrators' gaze behavior across a spectrum of demonstration devices. Our experimental results indicate that devices emulating (1) a robot's embodiment or (2) visual condition impair demonstrators' capability to extract task-relevant cues via gaze behavior, with the extent of impairment depending on the degree of emulation. Additionally, gaze data collected using devices that capture natural human behavior improves the policy's task success rate from 18.8% to 68.8% under environmental shifts.

Current invasive assistive technologies are designed to infer high-dimensional motor control signals from severely paralyzed patients. However, they face significant challenges, including public acceptance, limited longevity, and barriers to commercialization. Meanwhile, noninvasive alternatives often rely on artifact-prone signals, require lengthy user training, and struggle to deliver robust high-dimensional control for dexterous tasks. To address these issues, this study introduces a novel human-centered multimodal AI approach as intelligent compensatory mechanisms for lost motor functions that could potentially enable patients with severe paralysis to control high-dimensional assistive devices, such as dexterous robotic arms, using limited and noninvasive inputs. In contrast to the current state-of-the-art (SoTA) noninvasive approaches, our context-aware, multimodal shared-autonomy framework integrates deep reinforcement learning algorithms to blend limited low-dimensional user input with real-time environmental perception, enabling adaptive, dynamic, and intelligent interpretation of human intent for complex dexterous manipulation tasks, such as pick-and-place. The results from our ARAS (Adaptive Reinforcement learning for Amplification of limited inputs in Shared autonomy) trained with synthetic users over 50,000 computer simulation episodes demonstrated the first successful implementation of the proposed closed-loop human-in-the-loop paradigm, outperforming the SoTA shared autonomy algorithms. Following a zero-shot sim-to-real transfer, ARAS was evaluated on 23 human subjects, demonstrating high accuracy in dynamic intent detection and smooth, stable 3D trajectory control for dexterous pick-and-place tasks. ARAS user study achieved a high task success rate of 92.88%, with short completion times comparable to those of SoTA invasive assistive technologies.