Bipedal locomotion control is essential for humanoid robots to navigate complex, human-centric environments. While optimization-based control designs are popular for integrating sophisticated models of humanoid robots, they often require labor-intensive manual tuning. In this work, we address the challenges of parameter selection in bipedal locomotion control using DiffTune, a model-based autotuning method that leverages differential programming for efficient parameter learning. A major difficulty lies in balancing model fidelity with differentiability. We address this difficulty using a low-fidelity model for differentiability, enhanced by a Ground Reaction Force-and-Moment Network (GRFM-Net) to capture discrepancies between MPC commands and actual control effects. We validate the parameters learned by DiffTune with GRFM-Net in hardware experiments, which demonstrates the parameters’ optimality in a multi-objective setting compared with baseline parameters, reducing the total loss by up to 40.5% compared with the expert-tuned parameters. The results confirm the GRFM-Net’s effectiveness in mitigating the sim-to-real gap, improving the transferability of simulation-learned parameters to real hardware.

Autonomous visual navigation is an essential element in robot autonomy. Reinforcement learning (RL) offers a promising policy training paradigm. However, existing RL methods suffer from high sample complexity, poor sim-to-real transfer, and limited runtime adaptability. These problems are particularly challenging for drones, with complex nonlinear and unstable dynamics, and strong dynamic coupling between control and perception. In this paper, we propose a novel framework that integrates 3D Gaussian Splatting (3DGS) with differentiable deep reinforcement learning (DDRL) to train vision-based drone navigation policies. By leveraging high-fidelity 3D scene representations and differentiable simulation, our method improves sample efficiency and sim-to-real transfer. Additionally, we incorporate a Context-aided Estimator Network (CENet) to adapt to environmental variations at runtime. Moreover, by curriculum training in a mixture of different surrounding environments, we achieve in-task generalization, the ability to solve new instances of a task not seen during training. Drone hardware experiments demonstrate our method’s high training efficiency compared to state-of-the-art RL methods, zero shot sim-to-real transfer for real robot deployment without fine tuning, and ability to adapt to new instances within the same task class (e.g. to fly through a gate at different locations with different distractors in the environment). Our simulator and training framework are open-sourced at: https://github.com/Qianzhong-Chen/grad_nav.

As healthcare popularity increases, more people use food therapy for nourishment and healing. However, without scientific guidance, it's difficult to select appropriate foods for specific needs. To address the issue, we extract knowledge from TCMSP and professional books and fuse the data from different sources. Next, the Food Therapy Knowledge Graph (FTKG) is constructed. Finally, a food therapy system is developed that integrates the concept of TCMSP and FTKG, which uses the efficient knowledge retrieval and knowledge reasoning ability of the knowledge graph. It provides scientific food therapy solutions by analyzing symptoms and substituting traditional Chinese medicine with food, s address individual health needs.

Traditional Chinese medicine (TCM) has a unique advantage of preventive treatment of diseases, and adopting the concept of early intervention can effectively prevent diseases. Using knowledge graph is an effective way while the knowledge in the field of TCM is huge and messy. However, the structure of the TCM knowledge graph is often relatively sparse, which makes it highly limited. To this end, a rule-based compositional representation learning (RCRL) model is proposed. RCRL uses the implicit rules in the TCM knowledge graph, which solves the problem of poor representation learning due to the sparse structure of the TCM knowledge graph to a certain extent. Extensive experiments are conducted on the TCM knowledge graph and public datasets, and they are compared with other baselines. Experimental results show that RCRL is superior to other baselines, with improved learning accuracy and interpretability, and can be used for various downstream tasks.