Large language models (LLMs) demonstrate strong multilingual capabilities, yet often fail to consistently generate responses in the intended language, exhibiting a phenomenon known as language confusion.Prior mitigation approaches based on sequence-level fine-tuning, such as DPO, ORPO, and GRPO, operate at the level of entire responses and can lead to unintended degradation of general model capabilities, motivating the need for more fine-grained alternatives.To address this, we introduce Token-Level Policy Optimization (TLPO), a fine-tuning framework designed to mitigate language confusion through localized, token-level updates. TLPO identifies error-prone positions, explores alternative candidate tokens, and updates the policy using a tailored objective to suppress error-inducing outputs at a granular level.This selective intervention enables effective mitigation of language confusion without compromising the model’s general abilities.Experiments on multiple multilingual LLMs across diverse languages demonstrate that TLPO significantly outperforms baselines in improving language consistency while preserving downstream task accuracy.

Achieving robust performance is crucial when applying deep reinforcement learning (RL) in safety critical systems. Some of the state of the art approaches try to address the problem with adversarial agents, but these agents often require expert supervision to fine tune and prevent the adversary from becoming too challenging to the trainee agent. While other approaches involve automatically adjusting environment setups during training, they have been limited to simple environments where low-dimensional encodings can be used. Inspired by these approaches, we propose genetic curriculum, an algorithm that automatically identifies scenarios in which the agent currently fails and generates an associated curriculum to help the agent learn to solve the scenarios and acquire more robust behaviors. As a non-parametric optimizer, our approach uses a raw, non-fixed encoding of scenarios, reducing the need for expert supervision and allowing our algorithm to adapt to the changing performance of the agent. Our empirical studies show improvement in robustness over the existing state of the art algorithms, providing training curricula that result in agents being 2 - 8x times less likely to fail without sacrificing cumulative reward. We include an ablation study and share insights on why our algorithm outperforms prior approaches.

Most airdropped cargo use a combination of one or more parachutes and an impact attenuation system to land safely. The latter adds cost, weight and complexity. However, by using their legs for impact attenuation, airdropped quadruped robots may avoid the need for such a system. In this paper, various leg configurations for attenuating impact of airborne landings were studied and tested. Using simple lumped element models for simulation and analysis, a quadruped robot with a three-segment leg was designed and built. This model was validated with experiments with a small scale 20 cm-tall test robot. During the experiments, the test robot experienced 7.7 × 10 m/s2or 7.9 g-acceleration when dropped from height of 37.85 cm. This result is much better than the result of 1.4 × 102m/s2or 14.7g-acceleration when dropped at 10% of the original height with the same robot equipped with rigid legs. Such compliant leg design could be potentially used for impact attenuation of airdrop landings of robots five times larger.