Wenkai Chen

Robotic caregiving is emerging as a critical application domain where human-centered robotics, learning-based manipulation, and embodied intelligence converge. As global aging accelerates, assistive robots are expected to support essential daily activities, such as feeding, grooming, dressing, and body repositioning, which demand precise, adaptive, and safe physical interaction with users. Despite rapid progress, these tasks remain challenging due to deformable objects, diverse human behaviors, and safety-critical, contact-rich dynamics. This survey provides the first unified review of learning-based robotic caregiving across all four core manipulation tasks. We analyze recent advances through a cross-cutting framework grounded in four major learning paradigms: multimodal perception, imitation learning, reinforcement learning, and embodied reasoning. For each caregiving task, we synthesize how these paradigms address unique perceptual, physical, and safety requirements, identifying shared principles and task-specific characteristics. We also examine simulation platforms as enablers for scalable training and reproducible evaluation. This survey further summarizes a conceptual unified perception-reasoning-control loop that distills common structures across caregiving tasks. Finally, we highlight open challenges, including user-centered personalization, safety validation, and ethical deployment, and outline research directions toward trustworthy and adaptive assistive robots.

Existing task-oriented grasping approaches often rely on 2D pixel-wise affordance segmentation or predefined part annotations, limiting their applicability in unstructured 3D environments and constraining the grasp planning space. To overcome these limitations, we introduce a novel affordance-labeled grasp dataset constructed on simulation, capturing diverse functional interactions across object categories in a 6-DoF space. Building on this foundation, we propose a unified, language-guided grasping framework that takes partial point clouds and natural language instructions as input to generate semantically meaningful and geometrically feasible grasp poses. Specifically, a vision-language affordance grounding module produces dense 3D affordance maps aligned with task semantics, and a task-oriented grasp pipeline predicts coarse grasp candidates with implicit affordance cues. The coarse grasp proposals are subsequently refined based on visual affordance guidance, significantly enhancing both semantic alignment and grasp practicality. Extensive experiments in synthetic and real-world scenarios demonstrate that our method outperforms state-of-the-art approaches, effectively generalizing across diverse objects and tasks.

Accurate real-time prediction of diesel soot emissions is essential for meeting stringent emission regulations and optimizing aftertreatment performance. This study presents a data-driven framework for engine-out soot mass prediction, developed using experiments on an SC7H turbocharged diesel engine under the non-road transient cycle. A comprehensive dataset of soot concentration and engine sensor variables was analyzed using correlation-based network analysis, which identified diesel particulate filter (DPF) inlet oxygen concentration as a key factor in soot formation and oxidation. Two predictive models, a Light Gradient-Boosting Machine and a long short-term memory (LSTM) neural network, were trained to capture transient soot dynamics. A self-learning fusion strategy integrated their outputs into a meta-learner, forming an ensemble model. The fused model improved prediction accuracy by 7–10% over individual models and effectively tracked rapid emission fluctuations. This virtual sensing approach enables precise soot loading estimation, supporting optimal DPF regeneration timing and advanced emission control strategies, and offers a scalable methodology for environmentally sustainable diesel engine design in Industry 4.0.

Affordance understanding, the task of identifying actionable regions on 3D objects, plays a vital role in allowing robotic systems to engage with and operate within the physical world. Although Visual Language Models (VLMs) have excelled in high-level reasoning and long-horizon planning for robotic manipulation, they still fall short in grasping the nuanced physical properties required for effective human-robot interaction. In this paper, we introduce PAVLM (Point cloud Affordance Vision-Language Model), an innovative framework that utilizes the extensive multimodal knowledge embedded in pre-trained language models to enhance 3D affordance understanding of point cloud. PAVLM is an approach to integrates a geometric-guided propagation module with hidden embeddings from large language models (LLMs) to enrich visual semantics. On the language side, we prompt Llama-3.1 models to generate refined context-aware text, augmenting the instructional input with deeper semantic cues. Experimental results on the 3D-AffordanceNet benchmark demonstrate that PAVLM outperforms baseline methods for both full and partial point clouds, particularly excelling in its generalization to novel open-world affordance tasks of 3D objects. For more information, visit our project site: pavlm-source.github.io.

The capability for robotic systems to rearrange objects based on human instructions represents a critical step toward realizing embodied intelligence. Recently, diffusion-based learning has shown significant advancements in the field of data generation while prompt-based learning has proven effective in formulating robot manipulation strategies. However, prior solutions for robotic rearrangement have overlooked the significance of integrating human preferences and optimizing for rearrangement efficiency. Additionally, traditional prompt-based approaches struggle with complex, semantically meaningful rearrangement tasks without predefined target states for objects. To address these challenges, our work first introduces a comprehensive two dimensional (2-D) tabletop rearrangement dataset, utilizing a physical simulator to capture interobject relationships and semantic configurations. Then, we present DreamArrangement, a novel language-conditioned object rearrangement scheme, consisting of two primary processes: employing a transformer-based multimodal denoising diffusion model to envisage the desired arrangement of objects, and leveraging a vision–language foundational model to derive actionable policies from text, alongside initial and target visual information. In particular, we introduce an efficiency-oriented learning strategy to minimize the average motion distance of objects. Given few-shot instruction examples, the learned policy from our synthetic dataset can be transferred to the real world without extra human intervention. Extensive simulations validate DreamArrangement’s superior rearrangement quality and efficiency. Moreover, real-world robotic experiments confirm that our method can adeptly execute a range of challenging, language-conditioned, and long-horizon tasks with a singular model. The demonstration video can be found at https://youtu.be/fq25-DjrbQE

Visual servoing is a fundamental approach for robotic manipulation that relies on visual feedback to precisely control robot motion. Most methods are capable of generating velocity control signals to guide the camera to the desired position and orientation, which often exhibit limitations in dynamic responsiveness and robustness against noise and unmodeled dynamics. This paper presents an innovative acceleration-level position-based visual servoing control framework enhanced by deep reinforcement learning (DRL) integrated with Transformer-based temporal sequence processing. The essence of the method comprises two key elements: First, the controller retains the theoretical approach of position-based visual servoing in its design, ensuring transparency and a guaranteed performance baseline. Second, considering the temporal characteristics of servoing control, a Transformer-based actor-critic architecture within a Proximal Policy Optimization (PPO) reinforcement learning scheme is proposed to improve the learning efficiency and performance. Comprehensive experiments are conducted in both simulation and real robot scenarios. The results reveal that, compared with traditional velocity-level controllers, the proposed method demonstrates superior dynamic characteristics, enhanced tracking performance, and diminished sensitivity to noise in Cartesian space.

Robotic tool manipulation requires understanding task-relevant semantics under visually challenging conditions, such as shape variation and occlusion. This paper presents a novel framework for Language-Embedded Semantic 6D Pose Estimation that combines natural language instructions with 3D point cloud data to achieve category-level 6D pose estimation of tools' functional parts. By embedding semantic information from large language models (LLMs) and leveraging a diffusion-based pose estimator, our approach achieves robust generalization across diverse tool categories. We introduce a comprehensive synthetic dataset, tailored for tool manipulation scenarios, with annotated 6D poses of functional parts. Extensive experiments conducted on both the synthetic dataset and real-world robots demonstrate our system's ability to interpret natural language commands, predict poses of functional parts, and perform manipulation tasks with significant improvements in accuracy and generalization.

Affordance understanding, the task of identifying actionable regions on 3D objects, plays a vital role in allowing robotic systems to engage with and operate within the physical world. Although Visual Language Models (VLMs) have excelled in high-level reasoning and long-horizon planning for robotic manipulation, they still fall short in grasping the nuanced physical properties required for effective human-robot interaction. In this paper, we introduce PAVLM (Point cloud Affordance Vision-Language Model), an innovative framework that utilizes the extensive multimodal knowledge embedded in pre-trained language models to enhance 3D affordance understanding of point cloud. PAVLM integrates a geometric-guided propagation module with hidden embeddings from large language models (LLMs) to enrich visual semantics. On the language side, we prompt Llama-3.1 models to generate refined context-aware text, augmenting the instructional input with deeper semantic cues. Experimental results on the 3D-AffordanceNet benchmark demonstrate that PAVLM outperforms baseline methods for both full and partial point clouds, particularly excelling in its generalization to novel open-world affordance tasks of 3D objects. For more information, visit our project site: pavlm-source.github.io.