# Comment Here
# Comment Here
# Maybe question for audience if they have used ros2cli before. -# If they have, they are in the right place.
# Comment Here
# Comment Here
# Comment Here
# Comment Here
# Comment Here
# Comment Here
# These are just ideas, not even sure that is doable.
# Comment Here
# Comment Here
# I am not even sure what could be done with it. need to come back here to consider the possibility.
# Comment Here
# Comment Here
# Questions for audience like +# - if they have used ros2cli before? +# - if they have met the problem to figure out sub-command, parameter how to use those? +# If they have, they are in the right place. +# +# ros2ai is easy to use with container, just type `docker run` is all you need to do. +# +# Demo: +# - checking the status with openAI, `ros2 ai status -lv` to see API works okay and all the available models +# - 1st demo is query to ask the question about ROS 2, ros2ai can answer your question about ROS 2 based on your current distribution environment. +# - `ros2 ai query "what is the qos durability?"` +# - `ros2 ai query "how can we create the package"` +# - next is to ask AI to execute the appropriate command based on your request. ros2ai can execute the command behalf of you, and give the response back to you. that means you can iterate with ROS 2 system without figuring out the command line tools. +# - `ros2 ai exec "give me all the topic"` and we can check the details, `ros2 ai exec "detailed info about topic /rosout"` +# - it works with any languages, `ros2 ai exec "γγ©γ‘γΌγΏγͺγΉγγγγ γγ"` +# Demo shows us what we can do with ros2ai!
# Just introduce the motivation quickly with adding extra information. +# We are gonna talk about these with the following slides as a background including feature plan.
# this is what we do, but we do not want to if not necessary, so lets not do that. +# giving the example about QoS. +# before we get to know the QoS, we need to google, check the list, find appropriate site, and finally we still need to find the contents from ros2 documentation. +# this is expensive especially for beginners... nobody wants to do this.
# might sound like a strong opinion, but this is true. +# we obviously do not want to remember or know the every single command line interfaces, options and parameters. +# this is not only for the ROS 2 CLI but any systems. +# if would be really easier for us that someone interprets the communication between us and system. +# this would be great for maintainers, deprecated commands and options can be concealed by ros2ai.
# there has been several discussion brought up that we should support multi-language documentation. +# most OSS community does if the number of users are large such as CNCF projects for example. +# but, with my personal experience, every time i visit the multi-language documentation, in my case that is Japanese, there is a warning on the top saying "This is not up-to-dated, please visit English page." or something like that. +# this does not work at all, this system is already broken in the 1st place... +# instead of having the dedicated human resource to support this, my opinion is to let the AI work as we ask in their language based on the central precise information from the mainline. +# so that we can focus and pay attention on the mainline documentation with precise and concise information with quality.
# my focus is to keep it simple. +# implementation is really straight-forward using OpenAI Python API and ros2cli interfaces. +# unfortunately current ros2ai has sub-commands such as query and exec, but these need to be removed in the future. +# expecting this can be done with Function Calling feature that we are going to talk about in this presentation. +# i believe the most important thing is to keep is simple so that user can use without even thinking.
# ros2ai sits in between ROS 2 environment and OpenAI Python API like this. +# OpenAI Python API is compatible with Ollama, so that ros2ai can be agnostic from AI service via interface. +# you can configure backend AI service endpoint with using environmental variables. +# so every time you type the command in ros2ai, that is gonna process the request with OpenAI or Ollama, and iterate with ROS 2 system then comes back to you with result.
# As i introduced in the demonstration, we have now three sub-commands that are `status`, `query` and `exec`. +# if you query or execute the ros2ai, ros2ai fetches your ROS 2 environment setting such as what kind of distribution that you are using here, then integrate those information to the request completion field to AI backend service. +# that said, you can have the answer based on your local ROS 2 environment setting.
# Since ros2ai is new project and there is a huge area that we need to explore with using AI, let me explain about the future development and proposals in my mind.
# OpenAI has many parameters available to tune the behavior of AI. +# as user experience, i would pay attention about stable behavior and response time. +# it would not be useful if the behavior is not predictable or stable for user experience... +# and also response time should be reasonably quick enough to get the answer from ros2ai, if that is lagging, that could be frustration for user. +# besides these, i would like to explore to more base models and model information for better experience.
# for now, ros2ai does not support session cross-conversation result. +# that said, you can ask "/chatter topic available?", and ros2ai shows "/chatter", but you cannot ask "subscribe it". +# this is just because ros2ai is not aware of the session, so it does not understand this `it` for now. +# off the top of my head, it would be easy to support this with LangChain library to support the history just like other AI application does. +# but the question here is, history depth, where to cache, what the user interaction would be... those need to be well considered before implementation.
# For doing this ros2ai can act like a bridge between ROS 2 application and OpenAI or Ollama with text interfaces. +# this is totally doable, but expecting that application requirement could be different from the ros2ai. +# if that is the case, this would not be really that useful for other applications. +# at least, if we can spawn the daemon process by design, we can use ros2ai anywhere in the ROS 2 system, probably this can be useful for some cases.
# this is gonna be a good feature to support the most simple use of `ros2ai` without any options and sub-commands. +# currently it is the responsibility and burden for user to tell the sub-command category so that ros2ai can be told the context what needs to be done, question or execution? +# although this works for now, i believe that should be processed and categorized by AI, instead of user's burden. +# i think we can take advantage of Function Calling method to categorize the request with using AI, and then process the next user's request with that predicted context. +# i am expecting this would be useful to support more user friendly interfaces with ros2ai.
# we have 2 aspects here, the one is general tuning, and the other is user's specific data based tuning. +# Since this requires huge amount of time and cost, i am not even sure what we can do with this. +# besides, I am not sure if we can fetch the user's private environmental information to make it better. +# probably some users do not want to do this because of data privacy, but that is required to upload if user wants the AI to answer the your environment specific questions or requests. +# this would be something we are going to find out how we would want to proceed with community...
# i will be working with Robotec.AI as a member of ROS community to share experience and technical things. +# there will be AI WG, that is announced on this discourse thread. +# if you are interested, please feel free to join us and make difference together.