Catching objects in flight (i.e., thrown objects) is a common daily skill for humans, yet it presents a significant challenge for robots. This task requires a robot with agile and accurate motion, a large spatial workspace, and the ability to interact with diverse objects. In this paper, we build a mobile manipulator composed of a mobile base, a 6-DoF arm, and a 12-DoF dexterous hand to tackle such a challenging task. We propose a two-stage reinforcement learning framework to efficiently train a whole-body-control catching policy for this high-DoF system in simulation. The objects' throwing configurations, shapes, and sizes are randomized during training to enhance policy adaptivity to various trajectories and object characteristics in flight. The results show that our trained policy catches diverse objects with randomly thrown trajectories, at a high success rate of about 80 % in simulation, outperforming the baselines significantly. The policy trained in simulation can be deployed seamlessly to the real world with only onboard sensing and computation, which achieves catching sandbags in various shapes, randomly thrown by humans. Our video and code are available at https://mobile-dex-catch.github.io.

Behavior cloning has shown promise for robot manipulation, but real-world demonstrations are costly to acquire at scale. While simulated data offers a scalable alternative, particularly with advances in automated demonstration generation, transferring policies to the real world is hampered by various simulation and real domain gaps. In this work, we propose a unified sim-and-real co-training framework for learning generalizable manipulation policies that primarily leverages simulation and only requires a few real-world demonstrations. Central to our approach is learning a domain-invariant, task-relevant feature space. Our key insight is that aligning the joint distributions of observations and their corresponding actions across domains provides a richer signal than aligning observations (marginals) alone. We achieve this by embedding an Optimal Transport (OT)-inspired loss within the co-training framework, and extend this to an Unbalanced OT framework to handle the imbalance between abundant simulation data and limited real-world examples. We validate our method on challenging manipulation tasks, showing it can leverage abundant simulation data to achieve up to a 30\% improvement in the real-world success rate and even generalize to scenarios seen only in simulation.

Convolutional neural networks show their advantage in human attribute analysis (e.g. age, gender and ethnicity). However, they experience issues (e.g. robustness and responsiveness) when deployed in an intelligent video system. We propose one compact CNN model and apply it in our video system motivated by the full consideration of performance and usability. With the proposed web image mining and labelling strategy, we construct a large training set which covers various image conditions. The proposed CNN model successfully achieves a mean absolute error (MAE) of 3.23 years on the Morph 2 dataset, using the same test policy as our counterparts. This is the state-of-the-art score to our knowledge using CNN for age estimation. The proposed video analysis system employs this compact CNN model and demonstrated good performance in both dataset tests and deployment in real-world environments.

In this paper, TSV noise coupling is investigated in terms of both noise transfer function and signal integrity. An electrical model of coupled TSVs is presented and verified by 3-D field solver. Additionally, influence on TSV noise coupling transfer function with different TSV height, pitch and the thickness of insulation layer is analyzed. The electrical model helps to simplify the analyzing process and accurately reflect the influence of TSV process on coupled noise, thus providing a foundation for analysis on the noise-affected TSV signal. Further investigations on signal integrity of a TSV cluster are presented based on the electrical model, including transmission delay and bit error rate. The quantitative results tell an extensive influence on the reliability of TSV transmission system with different input patterns. The maximum transfer delay differs 10.24 times and signal quality differs up to several decades' times. The quantitative analysis in this paper can provide a theoretical foundation and insight of the anti-noise TSV design.