In Switzerland legal translation is uniquely important due to the country’s four official languages and requirements for multilingual legal documentation. However, this process traditionally relies on professionals who must be both legal experts and skilled translators—creating bottlenecks and impacting effective access to justice. To address this challenge, we introduce SwiLTra-Bench, a comprehensive multilingual benchmark of over 180K aligned Swiss legal translation pairs comprising laws, headnotes, and press releases across all Swiss languages along with English, designed to evaluate LLM-based translation systems. Our systematic evaluation reveals that frontier models achieve superior translation performance across all document types, while specialized translation systems excel specifically in laws but under-perform in headnotes. Through rigorous testing and human expert validation, we demonstrate that while fine-tuning open SLMs significantly improves their translation quality, they still lag behind the best zero-shot prompted frontier models such as Claude-3.5-Sonnet. Additionally, we present SwiLTra-Judge, a specialized LLM evaluation system that aligns best with human expert assessments.

Autocurricular training is an important sub-area of multi-agent reinforcement learning (MARL) that allows multiple agents to learn emergent skills in an unsupervised co-evolving scheme. The robotics community has experimented auto-curricular training with physically grounded problems, such as robust control and interactive manipulation tasks. However, the asymmetric nature of these tasks makes the generation of sophisticated policies challenging. Indeed, the asymmetry in the environment may implicitly or explicitly provide an advantage to a subset of agents which could, in turn, lead to a low-quality equilibrium. This paper proposes a novel game-theoretic algorithm, Stackelberg Multi-Agent Deep Deterministic Policy Gradient (ST-MADDPG), which formulates a two-player MARL problem as a Stackelberg game with one player as the ‘leader’ and the other as the ‘follower’ in a hierarchical interaction structure wherein the leader has an advantage. We first demonstrate that the leader's advantage from ST-MADDPG can be used to alleviate the inherent asymmetry in the environment. By exploiting the leader's advantage, ST-MADDPG improves the quality of a co-evolution process and results in more sophisticated and complex strategies that work well even against an unseen strong opponent.

The primary focus of this work is to examine how robots can achieve more robust sequential manipulation through the use of pre-touch sensors. The utility of close-range proximity sensing is evaluated through a robotic system that uses a new optical time-of-flight pre-touch sensor to complete a highly precise and sequential task - solving the Rubik's cube. The techniques used in this task are then extended to a more general framework in which ICP is used to match pre-touch data to a reference model, demonstrating that even simple pre-touch scans can be used to recover the pose of common objects that require sequential manipulation.

For certain manipulation tasks, object pose estimation from head-mounted cameras may not be sufficiently accurate. This is at least in part due to our inability to perfectly calibrate the coordinate frames of today's high degree of freedom robot arms that link the head to the end-effectors. We present a novel framework combining pre-touch sensing and deep learning to more accurately estimate pose in an efficient manner. The use of pre-touch sensing allows our method to localize the object directly with respect to the robot's end effector, thereby avoiding error caused by miscalibration of the arms. Instead of requiring the robot to scan the entire object with its pre-touch sensor, we use a deep neural network to detect object regions that contain distinctive geometric features. By focusing pre-touch sensing on these regions, the robot can more efficiently gather the information necessary to adjust its original pose estimate. Our region detection network was trained using a new dataset containing objects of widely varying geometries and has been labeled in a scalable fashion that is free from human bias. This dataset is applicable to any task that involves a pre-touch sensor gathering geometric information, and has been made publicly available. We evaluate our framework by having the robot re-estimate the pose of a number of objects of varying geometries. Compared to two simpler region proposal methods, we find that our deep neural network performs significantly better. In addition, we find that after a sequence of scans, objects can typically be localized to within 0.5 cm of their true position. We also observe that the original pose estimate can often be significantly improved after collecting a single quick scan.