Сознание как планировщик движения

[vak]

Читаю Метцингера и внутренне дискутирую с ним о причинах выработки эволюцией сознания у высших животных. Под сознанием понимается картина мира от первого лица, и интегральное ощущение себя в мире в момент времени "сейчас".

В принципе, почему бы не жить без всякого сознания. Можно заметить, что большая часть высших живых существ обходится без сознания. Имеются в виду растения и примкнувшие к ним грибы. Интересно, существуют ли бессознательные "зомби" среди животных? Для определённости, следи млекопитающих. Спросим искусственный интеллект.

Hi Gemini,
Do zombies exist in nature? I mean, mammals without conscience.

No, there aren't real-life zombies that are undead or lacking consciousness altogether.

However, there are some fascinating parasites and diseases that can alter an animal's behavior in dramatic ways, making them seem like zombies. Here are a couple of examples:

• Cordyceps Fungus: This fungus infects ants, taking control of their nervous system. The ant becomes a puppet, climbing a plant stem and then dying in a position that helps the fungus spread its spores.
• Rabies: This virus causes inflammation of the brain in mammals, including humans. Rabid animals can exhibit aggressive behavior and a strange hunger, though they aren't technically zombies.

So, while we don't have flesh-eating, undead creatures, nature has its own ways of creating "zombies" through parasites and diseases.

Ага, нету животных с отсутствующим сознанием. Только если мозг выели паразиты.

Но сознание дорогая вещь для организма. На его поддержание тратятся заметные жизненные ресурсы. Должен быть значительный эволюционный выигрыш от него, иначе зачем. И пришла мне в голову мысль, что сознание выполняет роль планирования движения для животного. Растениям без надобности, они никуда не спешат. А животному надо или догонять, или убегать, или с сексуальным партнёром заигрывать. И двигаться надо быстро, меняя решения за доли секунды. Картина от первого лица тут самое оно.

Сложности планирования движения хорошо известны инженерам, разрабатывающим автономных роботов. Снова поинтересуемся у Джемини.

What is the current best approach in industry and academia to motion planning for automotive robots? Can you give me an overview?

The current landscape of motion planning for automotive robots is a fascinating mix of established and cutting-edge techniques. Here's a breakdown of the popular approaches in both industry and academia:

Hybrid Deliberative/Reactive Architectures:

This is a well-established and widely used approach. It leverages a two-layered system:

• Deliberative Layer: This layer performs high-level, long-term planning using a map of the environment obtained from sensors like LiDAR.
• Reactive Layer: This layer handles real-time obstacles and unexpected situations.

This method offers a balance between global planning and adaptability to dynamic environments, making it a popular choice in industry for its reliability.

End-to-End Learning:

This is a growing area of research where the entire motion planning process is learned from data. Techniques like inverse reinforcement learning are being explored, where the robot learns from observing optimal behaviors. While promising, this approach might not be as readily deployable in industry due to potential safety concerns and the need for large datasets.

Differentiable Tree Policy Planning:

This is a recent advancement that builds upon traditional planning methods. It utilizes a tree structure to represent different motion possibilities and employs a special type of planning algorithm that can be continuously improved. This method offers advantages over single-stage planning by considering various options simultaneously and continuously refining them.

Industry vs. Academia:

• Industry: Prioritizes proven and reliable methods. Hybrid deliberative/reactive architectures are widely adopted due to their track record and ability to handle real-world complexities. Safety and predictability are paramount.
• Academia: Explores cutting-edge techniques like end-to-end learning and differentiable planning. The focus is on pushing the boundaries of performance and developing more efficient algorithms.

Overall, the field of motion planning for automotive robots is constantly evolving. While industry leans towards established, reliable methods, academia paves the way for future advancements. The hybrid approach seems to be the current sweet spot, but keep an eye on developments in end-to-end learning and differentiable planning as they mature.

Похоже, только простейшие и малоподвижные животные обходятся без сознания, типа червей и медуз. У остальных картина мира от первого лица, вероятнее всего, имеется.

Comments

[suhajh]

...более сложных систем, чем сам организм...
=вот бы мне свой перепланировать🤗