Title: Standing Tall: Robust Fall Prediction for Bipedal Robots Show Chinese title

Title: 挺直站立：双足机器人的鲁棒跌倒预测

Abstract: This paper extends the fall prediction algorithm from Mungai et al.(2024) to a real-time/online setting, implemented in both hardware and simulation. This yields results comparable to the offline version, maintaining a zero false positive rate, sufficient lead time, and accurate lead time prediction. Additionally, it achieves a high recovery rate. The paper also evaluates the fall prediction algorithm against omnidirectional faults and introduces an improved algorithm capable of reliably predicting falls and lead times across a wider range of faults in full-sized robots. Compared to Mungai et al.(2024), the proposed algorithm performs significantly better across all metrics, such as false positive rate, lead time, accuracy, and response time, demonstrating the algorithm's efficacy for real-time fall prediction in bipedal robots. Show Chinese abstract

Abstract: 本文将Mungai等人(2024)的跌倒预测算法扩展到实时/在线设置，在硬件和仿真中均进行了实现。这产生了与离线版本相当的结果，保持了零误报率、足够的提前时间以及准确的提前时间预测。此外，它还实现了高恢复率。本文还针对全向故障评估了跌倒预测算法，并引入了一种改进的算法，能够在更大范围的故障中可靠地预测跌倒和提前时间，适用于全尺寸机器人。与Mungai等人(2024)相比，所提出的算法在所有指标上表现显著更好，例如误报率、提前时间、准确性和响应时间，证明了该算法在双足机器人实时跌倒预测中的有效性。