About this Research Topic
This Research Topic is devoted to publishing extended versions of the papers presented in the Embodied Neuromorphic AI for Robotic Perception workshop held at the 2023 IEEE International Conference on Robotics and Automation (ICRA). All workshop contributions must be extended to include at least 30% original content. We also welcome relevant contributions that were not presented at the workshop.
The combination of neuromorphic circuits and sensor architectures is crucial for the development of a new generation of autonomous agents that have embodied neuromorphic AI. These agents can effectively interact with their environment, plan their actions, learn from their experiences, predict the outcomes of their actions, and adapt to changes in complex and uncertain environments. They can achieve these capabilities while using minimal resources like power, memory, and area overheads and while handling noise and variability. To achieve these goals, it is necessary to find solutions that enhance performance and robustness, which is different from the conventional engineering approach of adding general-purpose computing resources, redundancy, and control structures to the system. Despite remarkable progress in machine learning and computational neuroscience, conventional computing and robotics still cannot match human or animal performance in tasks that require embodied AI, like spatial perception and motor control.
Intelligent behaviour requires the processing of data on multiple timescales, which includes immediate perception analysis, hierarchical information extraction, and memorization of temporally structured data for long-term learning and adaptation. Although conventional computing can perform processes on different timescales, it results in power consumption and area/volume requirements that are far worse than those of biological neural networks. Neuromorphic engineering, on the other hand, uses mixed signal analogue/digital hardware that enables the implementation of neural computational primitives inspired by biological intelligence. This approach provides energy-efficient and compact solutions for implementing intelligence on robotic platforms. However, the research community faces several challenges in adopting this approach, including integrating full-custom neuromorphic chips with sensors, conventional computing modules, and motors, programming neural processing systems integrated on neuromorphic chips, and creating a principled framework for implementing and combining computational primitives, functions, and operations in these devices using neural instead of digital representations.
The design of robots that interact autonomously with the environment and exhibit complex behaviours is an open challenge that can benefit from understanding what makes living beings fit to act in the world. Neuromorphic engineering studies neural computational principles, developing technologies that can provide a computing substrate for building compact and low-power processing systems. In this workshop, we aim to discuss why endowing robots with neuromorphic technologies – from perception to motor control – represents a promising approach for the creation of robots which can seamlessly integrate into society. Highlighting open challenges in this direction, we propose community participation and actions required to overcome current limitations.
The combination of neuromorphic circuits and sensor architectures is crucial for the development of a new generation of autonomous agents that have embodied neuromorphic AI. These agents can effectively interact with their environment, plan their actions, learn from their experiences, predict the outcomes of their actions, and adapt to changes in complex and uncertain environments. They can achieve these capabilities while using minimal resources like power, memory, and area overheads and while handling noise and variability. To achieve these goals, it is necessary to find solutions that enhance performance and robustness, which is different from the conventional engineering approach of adding general-purpose computing resources, redundancy, and control structures to the system. Despite remarkable progress in machine learning and computational neuroscience, conventional computing and robotics still cannot match human or animal performance in tasks that require embodied AI, like spatial perception and motor control.
Intelligent behaviour requires the processing of data on multiple timescales, which includes immediate perception analysis, hierarchical information extraction, and memorization of temporally structured data for long-term learning and adaptation. Although conventional computing can perform processes on different timescales, it results in power consumption and area/volume requirements that are far worse than those of biological neural networks. Neuromorphic engineering, on the other hand, uses mixed signal analogue/digital hardware that enables the implementation of neural computational primitives inspired by biological intelligence. This approach provides energy-efficient and compact solutions for implementing intelligence on robotic platforms. However, the research community faces several challenges in adopting this approach, including integrating full-custom neuromorphic chips with sensors, conventional computing modules, and motors, programming neural processing systems integrated on neuromorphic chips, and creating a principled framework for implementing and combining computational primitives, functions, and operations in these devices using neural instead of digital representations.
The design of robots that interact autonomously with the environment and exhibit complex behaviours is an open challenge that can benefit from understanding what makes living beings fit to act in the world. Neuromorphic engineering studies neural computational principles, developing technologies that can provide a computing substrate for building compact and low-power processing systems. In this workshop, we aim to discuss why endowing robots with neuromorphic technologies – from perception to motor control – represents a promising approach for the creation of robots which can seamlessly integrate into society. Highlighting open challenges in this direction, we propose community participation and actions required to overcome current limitations.
Keywords: Embodied, Neuromorphic, AI, Artificial Intelligence, Robot, Perception
Important Note: All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.
