Search

Multi-legged robots offer enhanced stability in complex terrains, yet autonomously learning natural and robust motions in such environments remains challenging. Drawing inspiration from animals' progressive learning patterns, from simple to complex tasks, we introduce a universal two-stage learning framework with two-step reward setting based on self-acquired experience, which efficiently enables legged robots to incrementally learn natural and robust movements. In the first stage, robots learn through gait-related rewards to track velocity on flat terrain, acquiring natural, robust movements and generating effective motion experience data. In the second stage, mirroring animal learning from existing experiences, robots learn to navigate challenging terrains with natural and robust movements using adversarial imitation learning. To demonstrate our method's efficacy, we trained both quadruped robots and a hexapod robot, and the policy were successfully transferred to a physical quadruped robot GO1, which exhibited natural gait patterns and remarkable robustness in various terrains.

Learning multiple gaits is non-trivial for legged robots, especially when encountering different terrains and velocity commands. In this work, we present an end-to-end training framework for learning multiple gaits for quadruped robots, tailored to the needs of robust locomotion, agile locomotion, and user's commands. A latent space is constructed concurrently by a gait encoder and a gait generator, which helps the agent to reuse multiple gait skills to achieve adaptive gait behaviors. To learn natural behaviors for multiple gaits, we design gait-dependent rewards that are constructed explicitly from gait parameters and implicitly from conditional adversarial motion priors (CAMP). We demonstrate such multiple gaits control on a quadruped robot Go1 with only proprioceptive sensors.

Large pre-trained code generation models, such as OpenAI Codex, can generate syntax- and function-correct code, making the coding of programmers more productive and our pursuit of artificial general intelligence closer. In this paper, we introduce CodeGeeX, a multilingual model with 13 billion parameters for code generation. CodeGeeX is pre-trained on 850 billion tokens of 23 programming languages as of June 2022. Our extensive experiments suggest that CodeGeeX outperforms multilingual code models of similar scale for both the tasks of code generation and translation on HumanEval-X. Building upon HumanEval (Python only), we develop the HumanEval-X benchmark for evaluating multilingual models by hand-writing the solutions in C++, Java, JavaScript, and Go. In addition, we build CodeGeeX-based extensions on Visual Studio Code, JetBrains, and Cloud Studio, generating 4.7 billion tokens for tens of thousands of active users per week. Our user study demonstrates that CodeGeeX can help to increase coding efficiency for 83.4% of its users. Finally, CodeGeeX is publicly accessible and in Sep. 2022, we open-sourced its code, model weights (the version of 850B tokens), API, extensions, and HumanEval-X at https://github.com/THUDM/CodeGeeX.

We introduce GLM-130B, a bilingual (English and Chinese) pre-trained language model with 130 billion parameters. It is an attempt to open-source a 100B-scale model at least as good as GPT-3 (davinci) and unveil how models of such a scale can be successfully pre-trained. Over the course of this effort, we face numerous unexpected technical and engineering challenges, particularly on loss spikes and divergence. In this paper, we introduce the training process of GLM-130B including its design choices, training strategies for both efficiency and stability, and engineering efforts. The resultant GLM-130B model offers significant outperformance over GPT-3 175B (davinci) on a wide range of popular English benchmarks while the performance advantage is not observed in OPT-175B and BLOOM-176B. It also consistently and significantly outperforms ERNIE TITAN 3.0 260B -- the largest Chinese language model -- across related benchmarks. Finally, we leverage a unique scaling property of GLM-130B to reach INT4 quantization without post training, with almost no performance loss, making it the first among 100B-scale models and more importantly, allowing its effective inference on 4$\times$RTX 3090 (24G) or 8$\times$RTX 2080 Ti (11G) GPUs, the most affordable GPUs required for using 100B-scale models. The GLM-130B model weights are publicly accessible and its code, training logs, related toolkit, and lessons learned are open-sourced at \url{https://github.com/THUDM/GLM-130B/}., Comment: Accepted to ICLR 2023

We report CrISe monolayer as a room temperature ferromagnetic semiconductor with the Curie temperature ($T_C$), magnetic anisotropy energy (MAE) and band gap being 322 K, 113 $\mu$eV and 0.67 eV, respectively. The $T_C$ and MAE can be further enhanced up to 385 K and 313 $\mu$eV by tensile strain. More interestingly, the magnetic easy axis can be switched between off-plane and in-plane by compressive strain. Particularly, due to the broken inversion symmetry and strong spin-orbit coupling of Se atoms, a large Dzyaloshinskii-Moriya interaction (DMI) of 2.40 meV is obtained. More importantly, by micromagnetic simulations, stable skyrmions with sub-10 nm radius are stabilized by the large DMI above room temperature in a wide range of strain from $-2\%$ to $6\%$. Our work demonstrates CrISe as a promising candidate for next-generation skyrmion-based information storage devices and provides guidance for the research of DMI and skyrmions in room temperature ferromagnetic semiconductors., Comment: 12 pages, 5 figures

Theoretical prediction of Curie temperature (TC) is of vital importance for designing the spintronic devices in two-dimensional (2D) ferromagnetic materials. Herein, based on the extensive investigation of Monte Carlo simulations, we summary and propose an improved method to estimate TC more precisely, which includes the different contributions of multiple near-neighbor interactions. Taking monolayer CrI3 as an example, the trends of TC with biaxial strain are investigated via Monte Carlo simulations, mean-field formulas and our method. Besides, our method is not only accurate and convenient to predicting the TC in 2D ferromagnetic honeycomb lattice CrI3 but it can be extended for predicting the TC of other 2D lattices. Our work paves the way to accelerate the prediction and discovery of novel 2D ferromagnets for spintronic applications., Comment: 10 pages, 5 figures

Recently, great effort has been devoted to the search for two-dimensional (2D) ferromagnetic materials with inherent strong Dzyaloshinskii--Moriya interaction (DMI). Here, through a first-principles approach, we systematically investigate the effect of biaxial strain on the DMI, the Heisenberg exchange interaction, and the magnetic anisotropy energy (MAE) of Janus Cr$_{2}$X$_{3}$Y$_{3}$ (X, Y = Cl, Br, I, X $\neq$ Y) monolayers. Both DMI and MAE can be significantly enhanced by tensile strain, while a reversal of the chirality of DMI in Cr$_{2}$Cl$_{3}$Br$_{3}$ and a switch of MAE from off-plane to in-plane in Cr$_{2}$I$_{3}$Cl$_{3}$ are induced by a compressive strain of $2\%$. Microscopically, DMI and MAE are associated mainly with the large spin--orbit coupling of the heavy nonmagnetic halogen atoms rather than that of the magnetic Cr atoms. In particular, the peculiar magnetic transition of Cr$_{2}$I$_{3}$Cl$_{3}$ is explained by competition between direct exchange and superexchange interactions. Micromagnetic simulations show that a small external magnetic field of 65~mT stabilizes a skyrmion with a diameter of 9.8~nm in the Cr$_2$I$_3$Cl$_3$ monolayer. Our results will provide guidance for further research on DMI and skyrmions in 2D Janus materials, as well as a basis for the potential applications in spintronic devices., Comment: 8 pages, 5 figures