Tool-Integrated Reasoning has emerged as a key paradigm to augment Large Language Models (LLMs) with computational capabilities, yet integrating tool-use into long Chainof-Thought (long CoT) remains underexplored, largely due to the scarcity of training data and the challenge of integrating tool-use without compromising the model's intrinsic longchain reasoning.In this paper, we introduce DART (Discovery And Reinforcement of Tool-Integrated Reasoning Chains via Rollout Trees), a reinforcement learning framework that enables spontaneous tool-use during long CoT reasoning without additional human annotation.DART operates by constructing dynamic rollout trees during training to discover valid tool-use opportunities, branching out at promising positions to explore tool-integrated trajectories.Subsequently, a tree-based process advantage estimation identifies and credits specific sub-trajectories where tool invocation positively contributes to the solution, effectively reinforcing these beneficial behaviors during training.Extensive experiments on challenging benchmarks like AIME and GPQA-Diamond demonstrate that DART significantly outperforms existing methods, successfully harmonizing tool execution with long CoT reasoning.

With the rise of Speech Large Language Models (SpeechLLMs), two dominant approaches have emerged for speech processing: discrete tokens and continuous features.Each approach has demonstrated strong capabilities in audiorelated processing tasks.However, the performance gap between these two paradigms has not been thoroughly explored.To address this gap, we present a fair comparison of selfsupervised learning (SSL)-based discrete and continuous features under the same experimental settings.We evaluate their performance across six spoken language understandingrelated tasks using both small and large-scale LLMs (Qwen1.5-0.5B and Llama3.1-8B).We further conduct in-depth analyses, including efficient comparison, SSL layer analysis, LLM layer analysis, and robustness comparison.Our findings reveal that continuous features generally outperform discrete tokens in various tasks.Each speech processing method exhibits distinct characteristics and patterns in how it learns and processes speech information.We hope our findings will provide valuable insights to advance spoken language understanding in SpeechLLMs.

Cleaning high-altitude building windows is traditionally a high-risk task for human workers due to the danger of falls. To enhance safety in facade cleaning and similar tasks, the development of window-cleaning robots has become increasingly essential. However, current robots face challenges in navigating exoskeleton structures due to obstructive beams, limiting their application. This article proposes a novel biped window-cleaning robotic system (BWCRS) designed for exoskeleton structures. The system comprises two key components: a multifunctional biped climbing robot (BCR) and a specialized cleaning robot. The BCR is adept at overcoming obstacles inherent in exoskeleton structures and transitioning across various inclined window surfaces. The BCR adopts a symmetrical construction, and the ends of the feet are equipped with vacuum suction cups and electromagnets, which can switch between different actuators at the geometric base and end of the system. The system is further enhanced by air pressure sensors that monitor the suction cup adhesion in real time, ensuring stable operation. A crucial functionality of the BCR is its manipulation mode, guided by a visual camera. This mode enables precise location and manipulation of the cleaning robot, ensuring effective coverage of all window areas. The mechanical design and mathematical modeling of the BWCRS are thoroughly examined, demonstrating its robust window-cleaning capabilities. The experimental results reveal that the proposed BCR has a favorable window-climbing and manipulation ability and that the BWCRS has a favorable cleaning effect on the exoskeleton-structure windows. Note to Practitioners—This paper is motivated by the problem of window-cleaning for exoskeleton structures, but it also applies to solving window maintenance and inspection. Previous window-cleaning robots were mainly aimed at the facades or surfaces with low obstacles that cannot be applied directly to exoskeleton-structured windows. This paper proposes an exoskeleton-structure window-cleaning method based on a biped window-climbing robot and a cleaning robot. The biped robot can climb over obstacles on the exoskeleton structures and manipulate the cleaning robot to different cleaning areas. Here, we introduce the biped window-climbing robot from the perspective of mechanical design and mathematical kinematics. We then demonstrate multiple obstacle-crossing motion modes of the biped window-climbing robot. Combined with visual-based manipulation, the window-cleaning process for the exoskeleton structures by a biped window-climbing robot integrated with a cleaning robot is demonstrated. The results verify that the proposed design can complete window-cleaning tasks on exoskeleton structures.

Autonomous magnetic catheterization has become a promising technology for next-generation minimally invasive cardiovascular surgery. In this paper, a novel homocentric variable-stiffness magnetic robotic catheter (HVS-MRC) is presented. This device is controlled by a multi-arm robot-assisted catheterization system for use in radiation-free coronary ultrasound intervention. The uniqueness of the HVS-MRC is the generation of variable stiffness through the telescopic motion of three homocentric components with embedded internal magnets. These components can be used to achieve multiple curvatures and large deflections ($>$120 degrees) under a magnetic wrench provided by an external mobile magnet module (EMMM) to adapt to the complex coronary environment. The variable-stiffness kinematics modeling, multivessel selection strategy, and preoperative surgical planning of the proposed magnetic robotic catheter are established. Meanwhile, an extracorporeal mobile ultrasound module (EMUM) is adopted to intraoperatively localize the catheter's in-plane motion with a dominant visual/force feedback controller. A hierarchical relative control scheme is proposed based on the relative Jacobian method to synchronize the motions of the multi-arm robot-assisted catheterization system. The overall performance is evaluated with anin vitrohuman-sized coronary arterial phantom with challenging anatomical variabilities. The results reveal that the HVS-MRC exhibits a high-accuracy intervention performance (average error of 1.52$ \pm$0.35 mm) with smoother steering compared with conventional catheters. High synchronization with a low ultrasound target loss rate (15.8%) and constant-force tracing (2.50$\pm$1.02 N) of the multi-arm robot-assisted catheterization system demonstrate promising application potential in radiation-free autonomous ultrasonographic coronary intervention operations.

With the global population aging rapidly, Alzheimer’s disease (AD) is particularly prominent in older adults, which has an insidious onset and leads to a gradual, irreversible deterioration in cognitive domains (memory, communication, etc.). Speech-based AD detection opens up the possibility of widespread screening and timely disease intervention. Recent advances in pre-trained models motivate AD detection modeling to shift from low-level features to high-level representations. This paper presents several efficient methods to extract better AD-related cues from high-level acoustic and linguistic features. Based on these features, the paper also proposes a novel task-oriented approach by modeling the relationship between the participants’ description and the cognitive task. Experiments are carried out on the ADReSS dataset in a binary classification setup, and models are evaluated on the unseen test set. Results and comparison with recent literature demonstrate the efficiency and superior performance of proposed acoustic, linguistic and task-oriented methods. The findings also show the importance of semantic and syntactic information, and feasibility of automation and generalization with the promising audio-only and task-oriented methods for the AD detection task.

24th Annual Conference of the International Speech Communication Association, INTERSPEECH 2023, Dublin, Ireland, August 20-24, 2023