Compared to traditional data-driven learning methods, recently developed deep reinforcement learning (DRL) approaches can be employed to train robot agents to obtain control policies with appealing performance. However, learning control policies for real-world robots through DRL is costly and cumbersome. A promising alternative is to train policies in simulated environments and transfer the learned policies to real-world scenarios. Unfortunately, due to the reality gap between simulated and real-world environments, the policies learned in simulated environments often cannot be generalized well to the real world. Bridging the reality gap is still a challenging problem. In this paper, we propose a novel real–sim–real (RSR) transfer method that includes a real-to-sim training phase and a sim-to-real inference phase. In the real-to-sim training phase, a task-relevant simulated environment is constructed based on semantic information of the real-world scenario and coordinate transformation, and then a policy is trained with the DRL method in the built simulated environment. In the sim-to-real inference phase, the learned policy is directly applied to control the robot in real-world scenarios without any real-world data. Experimental results in two different robot control tasks show that the proposed RSR method can train skill policies with high generalization performance and significantly low training costs.

Visuotactile sensors have significant advantages in high-resolution contact shape sensing, which is crucial for robots to locate and manipulate objects. However, high precision and robust 3-D geometry sensing is still a challenging problem. In this article, we propose an improved GelStereo sensor, named GelStereo 2.0, with a more stable and compact structure. A refractive stereo ray tracing modeling method is proposed to eliminate 3-D geometry reconstruction errors caused by multimedium light refraction. In addition, we introduce a novel calibration method named multiple contact depths to obtain refractive parameters, which only needs a few pressings using calibration boards. Extensive experimental results show that the measurement error of contact surface depth is less than 0.15 mm. Furthermore, two well-calibrated sensors mounted on the gripper are successfully applied to a fingertip bolt screwing task, which demonstrates the effectiveness of the GelStereo 2.0 sensor and the proposed tactile 3-D geometry sensing method.

In this letter, we propose a novel Grasp Pose Domain Adaptation Network (GPDAN) to achieve sim-to-real domain adaptation for 6-DoF grasp pose detection. The main task of GPDAN is to detect feasible 6-DoF grasp poses in cluttered scenes. A point-wise self-supervised domain classification module with point cloud mixture and feature fusion strategy is proposed as the auxiliary task to promote the feature alignment between the source and target domain through adversarial training. Experimental results on both simulation and real-world environments demonstrate that GPDAN outperforms other approaches in detecting 6-DoF grasps on the target domain, highlighting the effectiveness of GPDAN in improving the performance of 6-DoF grasp pose detectors trained in simulation and deployed in real-world environments without any further laborious labeling.

Open circuit voltage (OCV) is crucial for battery degradation analysis. However, high-precision OCV is usually obtained offline. To this end, this paper proposes a novel self-evaluation criterion based on the capacity difference of State of Charge (SoC) unit interval. The criterion is integrated into extended Kalman filter (EKF) for joint estimations of OCV and SoC. The proposed method is evaluated in a typical application scenario, energy storage system (ESS), using a ${\mathrm{LiFePO}}_4$ (LFP) battery. Extensive experimental results show that a more accurate OCV and incremental capacity and differential voltage (IC-DV) can be achieved online with the proposed method. Our method also greatly improves the accuracy of SoC estimation at each SoC point where the maximum estimation error of SoC is less than 0.3%.

Long-horizon dexterous tasks have been a long-standing problem in robotic manipulation. Previous studies have developed task and motion planning, imitation learning, and reinforcement learning methods for long-horizon manipulations. However, these methods are hard to achieve efficient planning for new tasks. Empowered with the Vision Language Model (VLM), recent studies significantly improve the generalization of robot systems. However, these works are only verified in simple pick-and-place tasks due to limited skills. To this end, we propose the Learn-Gen-Plan (LGP), which combines the VLM and learning-based primitives to endow robots with the ability to efficiently plan and complete various long-horizon dexterous tasks. LGP contains two key phases: skill generation and task planning. In skill generation, the Skill Generator is proposed to utilize the learned key primitives and hand-crafted trivial primitives to generate adaptive robot skills. In task planning, the Multimodal Planner generates the robot plan based on image observation, generated skills, and text prompts. We set up a series of dexterous tasks (e.g., cable routing, peg-in-hole assembly) in a real-world lighting circuit wiring scenario to evaluate LGP. The experimental results show that LGP efficiently generates robot plans with learned skills, controlling the robot to complete various multi-step cable wiring tasks.

Introduction: In recent years, the perceptual capabilities of robots have been significantly enhanced. However, the task execution of the robots still lacks adaptive capabilities in unstructured and dynamic environments.Methods: In this paper, we propose an ontology based autonomous robot task processing framework (ARTProF), to improve the robot's adaptability within unstructured and dynamic environments. ARTProF unifies ontological knowledge representation, reasoning, and autonomous task planning and execution into a single framework. The interface between the knowledge base and neural network-based object detection is first introduced in ARTProF to improve the robot's perception capabilities. A knowledge-driven manipulation operator based on Robot Operating System (ROS) is then designed to facilitate the interaction between the knowledge base and the robot's primitive actions. Additionally, an operation similarity model is proposed to endow the robot with the ability to generalize to novel objects. Finally, a dynamic task planning algorithm, leveraging ontological knowledge, equips the robot with adaptability to execute tasks in unstructured and dynamic environments.Results: Experimental results on real-world scenarios and simulations demonstrate the effectiveness and efficiency of the proposed ARTProF framework.Discussion: In future work, we will focus on refining the ARTProF framework by integrating neurosymbolic inference.