In Evolutionary robotics, techniques inspired by biological evolution are used to evolve robotic morphology and behaviors. Evolution Strategies is a popular algorithm in this field due to its capability to perform con...
详细信息
ISBN:
(纸本)9798400701207
In Evolutionary robotics, techniques inspired by biological evolution are used to evolve robotic morphology and behaviors. Evolution Strategies is a popular algorithm in this field due to its capability to perform continuous optimization. In evolutionary robotics, the use of approaches as, for instance, niching techniques and environmental variation, has shown promising results in order to develop more effective adaptive behaviors or even improve the convergence time. The objective of this work is to analyze how a recent adaptation of Evolution Strategies developed by OpenAI, which we call OpenAI-ES, behaves when working alongside environmental variation and niching techniques. The results showed that this approach applied to the double pole balancing benchmark can improve performance by 9% for OpenAI-ES.
adaptive robotics plays an essential role in achieving truly co-creative cyber physical systems. In robotic manipulation tasks, one of the biggest challenges is to estimate the pose of given workpieces. Even though th...
详细信息
ISBN:
(纸本)9781713872344
adaptive robotics plays an essential role in achieving truly co-creative cyber physical systems. In robotic manipulation tasks, one of the biggest challenges is to estimate the pose of given workpieces. Even though the recent deep-learning-based models show promising results, they require an immense dataset for training. In this paper, two vision-based, multi-object grasp pose estimation models (MOGPE), the MOGPE Real-Time and the MOGPE High-Precision are proposed. Furthermore, a sim2real method based on domain randomization to diminish the reality gap and overcome the data shortage. Our methods yielded an 80% and a 96.67% success rate in a real-world robotic pick-and-place experiment, with the MOGPE Real-Time and the MOGPE High-Precision model respectively. Our framework provides an industrial tool for fast data generation and model training and requires minimal domain-specific data. Copyright (c) 2023 The Authors. This is an open access article under the CC BY-NC-ND license (https://***/licenses/by-nc-nd/4.0/)
In order to complete useful tasks in complex and changing contexts, robots need to be able to adapt their behavior or actions. This requires the ability to find effective approaches to the situation at hand, and to ma...
详细信息
In order to complete useful tasks in complex and changing contexts, robots need to be able to adapt their behavior or actions. This requires the ability to find effective approaches to the situation at hand, and to maintain them until they keep being effective. The final goal can be summarized by the will of maintaining a high performance during the whole life of the robot, regardless of the changes that may intervene during its activity. In this work, we evaluate by the point of view of the resulting life-long performance two methodologies for the adaptation of robots controlled by immutable network-based control systems: the Nanowire Networks. We demonstrate that modifying the best found solution leads to constant improvement and to overall better cumulative performance. Complementarily, we show that a less constrained approach simplifies the exploration of different behaviors but reduces life-long performance. Finally, we confirm previous results suggesting the potential of using this novel neuromorphic device (i.e., the Nanowire Network) for the control of robots.
adaptive robotics plays an essential role in achieving truly co-creative cyber physical systems. In robotic manipulation tasks, one of the biggest challenges is to estimate the pose of given workpieces. Even though th...
详细信息
adaptive robotics plays an essential role in achieving truly co-creative cyber physical systems. In robotic manipulation tasks, one of the biggest challenges is to estimate the pose of given workpieces. Even though the recent deep-learning-based models show promising results, they require an immense dataset for training. In this paper, two vision-based, multi-object grasp pose estimation models (MOGPE), the MOGPE Real-Time and the MOGPE High-Precision are proposed. Furthermore, a sim2real method based on domain randomization to diminish the reality gap and overcome the data shortage. Our methods yielded an 80% and a 96.67% success rate in a real-world robotic pick-and-place experiment, with the MOGPE Real-Time and the MOGPE High-Precision model respectively. Our framework provides an industrial tool for fast data generation and model training and requires minimal domain-specific data.
暂无评论