ChatGTP, a language-based chatbot developed by OpenAI, has gained a lot of use and attention since its launch in 2022. Since then, it and other chatbots – programs designed to mimic human language and conversation – have become increasingly common, used to provide content creation, marketing, programming, research, entertainment, customer service and more.
Embedded systems designs are also affected by the rapid development of artificial intelligence. Microcontrollers have limited processing power and memory – both of which are important for the speed, efficiency, and effectiveness of chatbots. Microcontrollers are typically not ideal for running advanced AI models that power most chatbots. But this is changing.
While running a chatbot directly on a microcontroller is still a niche concept, it is possible to overcome some of its limitations. In this article, we will discuss the use of chatbots in embedded applications, as well as the methods used to run chatbots with microcontrollers or microcomputer-based applications.
What are chatbots?
Chatbots are computer programs that simulate human conversations as text or audio. They are designed to interpret and respond to user queries, requests or commands. Chatbots have become excellent at interacting with users to provide information or complete a digital task.
Currently, there are three types of chatbot technologies.
1. Rules-based Chatbots: Follow a predefined set of rules and keywords to generate responses.
2. AI-powered chatbots: Use natural language processing (NLP) and machine learning to understand user intent and provide dynamic responses.
3. Hybrid chatbots: Use rules-based and AI approaches for greater flexibility.
Chatbots that work like computer programs can provide assistance or services 24/7, anytime, anywhere. They are useful for automating tasks, user interaction, and enhancing user experiences with technology.
Applications in embedded systems
Chatbots provide a useful means to implement NLP in embedded applications, improving user interactions through language.
Here are some typical chatbot applications with microcontrollers
Voice command recognition: One of the most common uses of chatbots in embedded devices, now frequently used in smart home systems, IoT devices, and automotive applications. It allows users to control, interact with, and configure devices through spoken language (such as setting an alarm or asking Siri to play music with an audio command). This is commonly applied in smart home systems, IoT devices, and automotive applications.
Troubleshooting and assistance: Chatbots can diagnose common issues with smart devices and guide users through troubleshooting steps. They can also connect users with human support personnel when needed.
Personalization: Chatbots serve as an active means of interacting with users. AI-based chatbots can “learn” user preferences. They can automate embedded systems or adjust smart device settings according to the user's usage pattern and behavior. This can lead to greater energy efficiency in homes, automatically set appliances (such as coffee makers) or alarms (at home or office) for convenience and security, and improve overall user comfort (by adjusting thermostats or other devices as desired).
Conversational Interfaces: Chatbots with NLP capabilities enable embedded smart devices to talk and interact with users. This is useful for customer service kiosks, interactive displays, and information retrieval systems.
Machine interface: Along with retrieving or executing tasks through user communication, chatbots can simplify interactions with complex industrial machines. Machine operators can use chatbot-powered interfaces to monitor conditions, issue commands, receive diagnostic information, or authorize access to critical machines.
Training and support: Chatbots have become useful in on-the-job training in some industries. Embedded devices powered by chatbots can explain certain tasks or procedures, teach users specific operational steps, provide real-time assistance or support with new equipment, and help with troubleshooting.
Predictive maintenance: Chatbots can detect potential equipment failures by analyzing sensor data and alerting maintenance personnel before failures occur, minimizing downtime. Because chatbots can operate 24/7, they can continuously monitor equipment and potentially resolve issues or suggest solutions for improvements.
Language translation: Chatbots can provide language translation services (in text or audio format) in real-time using NLP, supporting users while traveling or learning a new language.
Text summary: can summarize large volumes of text, making it easier for users to consume relevant information. This can be useful in embedded systems that support textual data, such as news readers or document summarizers. This is also useful in IoT applications that process and analyze large volumes of sensor data.
Reminders and notifications: Smart gadgets with chatbots can be programmed to generate notifications and alerts. For example, they can remind users of appointments or medication, generating warnings in natural language.
Emergency response: Chatbots in wearable gadgets can provide emergency response and assistance as needed, such as contacting emergency services, guiding users through first aid procedures, or providing emotional support.
Sentiment Analysis: A tool that embedded systems can employ to interpret and react to emotional responses from human input. This applies to social robots, feedback analysis systems or emotion-sensitive devices.
Gesture and language integration: Supports an embedded system to interpret spoken commands and gestures to control a device or system.
Text-based search and retrieval: NLP-driven search functions allow users to interact with embedded systems using natural language queries. This benefits applications such as intelligent search engines, content retrieval systems, and knowledge-based interfaces.
Health and fitness tracking: Chatbots can provide personalized feedback on activity levels, suggest exercise routines based on certain goals, and answer questions about health data.
Text-to-speech (TTS) and speech-to-text (STT): Using AI-based chatbots, embedded systems can convert spoken language to written text (STT) and vice versa (TTS). This is useful in applications such as voice assistants, accessibility tools, and communication devices.
Interaction with robots: supporting robots to “understand” human commands and intentions, leading to more intuitive and natural interactions.
Interactive educational tools: Chatbots on educational devices can facilitate natural language interactions for learning purposes. They can guide students in educational activities, tell stories, play games, or make robot companions more engaging and interactive. Chatbots are also useful as language tutors or language learning tools.
Security and access control: Used to increase security by implementing voice-based access control systems. Users can use voice as a secure authentication method for embedded devices.
Accessibility and assistance : Chatbots can help people with disabilities control robots using voice commands, assist with daily activities, or provide companionship.
Running chatbots with microcontrollers
It is not possible to run chatbots directly on microcontrollers, although many can run smaller LLM models and custom frameworks for high-end devices. However, alternative approaches are used to implement chatbots in most embedded and IoT applications.
Here are some practices for implementing chatbots with microcontrollers.
Rules-based chatbots: Rely on a set of predefined rules and keywords to respond to user input. These lightweight chatbots can run on basic microcontrollers, providing basic NLP capabilities in a smart device. However, interactions with users are limited and lack flexibility. They also lack the natural language understanding of AI-powered chatbots.
Cloud-based chatbots: In these chatbots, complex natural language processing is offloaded to a cloud-based chatbot platform. The microcontroller only connects the sensors and actuators, collecting user information and sending it to the cloud for processing. A cloud-based chatbot generates a response, which is sent back to the microcontroller and communicated via voice or text. In this method, the NLP functionality of the embedded system depends on the internet connectivity and the efficiency of the cloud platform. However, it incorporates natural language and allows greater flexibility in user interactions.
Offline keyword recognition with limited responses: Libraries like TinyML are used to train small keyword recognition models on the microcontroller. This model triggers pre-recorded responses or actions based on recognized keywords. This approach is useful in simple applications such as voice-controlled devices, enabling NLP functionalities without complex system and software design.
Chatbots for microcontroller
Below are some chatbot options for implementing NLP functions in microcontroller-based systems.
ChatGPT in the Cloud: Several embedded applications incorporate NLP by integrating cloud-based services that run ChatGPT. The microcontroller captures user input and sends it to a ChatGPT-based cloud service via an API. Responses are received by the online service and displayed by the device's microcontroller. For example, a microcontroller can retrieve user input via a keyboard or numeric keypad and send it to a remote server for processing. Then, the server processes the request and sends a response to the microcontroller, which communicates via voice or display.
Microsoft Azure Bot Service: A popular cloud-based chatbot that integrates with multiple channels, such as voice assistants, messaging apps, or web interfaces, giving users flexibility in how they interact with their devices. This platform also features design bots with buttons, images, and carousels to provide visually appealing and interactive experiences on an embedded device's interface.
The bot can track past interactions and adapt responses accordingly, offering a more personalized and contextual experience. It allows users to access device information, control settings, or trigger actions through natural language commands. A device can also delegate complex tasks or workflows to the bot, freeing up resources for core functionality. It can manage devices from a central location and ensures that all data communication is secure and private.
Dialog Flow: Google's cloud-based NLP platform is useful for embedded applications, depending on the project. It offers lightweight Dialogflow Agents (LDFAs) for resource-constrained devices. Users can combine cloud-based and on-device functionality for flexibility and optimal performance.
Dialogflow facilitates offline speech recognition engines to handle user input without internet connectivity. It is suitable for cloud-based deployment, although its agents can be used offline. It offers pre-built integrations with other Google services like Google Assistant, Calendar, and Maps that expand a bot's capabilities and enable personalized conversational flows. It tracks past interactions and adapts responses accordingly, providing a personalized and contextual experience. It has a user-friendly interface with many drag and drop tools, ideally requiring knowledge of Dialogflow and its related development tools.
ChatScript: A free, open-source, rules-based chatbot engine with extensive documentation and a large community base. It can work efficiently on any low-power device due to its rule-based approach. It is flexible in scripting language and allows you to adapt the chatbot's responses and behavior to the specific needs and scenarios of IoT and embedded systems. It works effectively even without an internet connection. However, it cannot deal with complex language, ambiguities and nuances. It also lacks built-in integrations, context awareness, and advanced analytics features.
Managing complex rules and logic to handle diverse user interactions is a challenge for a rules-based approach. ChatScript is suitable for incorporating simple NLP functions into embedded systems and IoT projects that do not rely on Internet connectivity.
AIML: stands for Artificial Intelligence Markup Language, a popular rules-based chatbot language based on simplicity and ease of use. It is ideal for integration with simple educational toys, voice interaction on embedded devices, and testing simple proof-of-concept prototypes such as voice interaction functionality. AIML relies on pattern matching and predefined rules. The use of AIML is declining as AI improves due to its lack of built-in functionalities such as context awareness, sentiment analysis, personalization, security concerns, and poor scalability.
Edge Boost: A platform for developing and deploying TinyML models on microcontrollers and IoT devices. It offers a user-friendly interface with many drag-and-drop tools and visual workflows and integrates with multiple sensors. Users can train and run machine learning models directly on the device, eliminating the need for cloud processing and reducing latency. It automatically optimizes models for size and efficiency, ensuring they run smoothly on microcontrollers. Various machine learning models such as keyword detection, anomaly detection, predictive maintenance, gesture recognition, and AI functions can be deployed without any hassle. This low-cost, flexible and efficient platform is best used for simpler tasks.
Tensorflow Lite Micro: A lightweight version of TensorFlow useful for deploying small conversational models on microcontrollers. The models run directly on the device and can operate without an internet connection. This reduces the power consumption of the embedded system. Tensorflow Lite is useful for battery-powered IoT devices and microcontrollers with limited resources. Developers can choose from a variety of pre-trained models or train themselves, tailoring functionality to their specific needs and hardware platform. It excels at simpler tasks and may not be suitable for running complex machine learning or deep learning models. Supports many popular platforms, but not all microcontrollers and hardware platforms.
Rasa NLU: A powerful open source framework for natural language understanding (NLU), enabling machines to extract meaning and intent from conversations. It is possible to run Rasa NLU on a Raspberry Pi or similar devices and communicate with it using a microcontroller.
Rasa excels at understanding the meaning and intent behind user queries, even with limited vocabulary or ambiguous phrases typically used for conversational AI for devices. Custom NLU models can be trained in the language and terminology of a user's specific domain, ensuring accurate understanding of intent. It integrates seamlessly with other chatbot platforms and conversational AI tools and offers multilingual support. Most Rasa NLU models require an Internet connection for processing.
ChatterBot: is a Python library for creating chatbots. It can be used to implement a chatbot on a microcomputer like Raspberry Pi and communicate with it from a microcontroller. It enables natural language interactions with embedded and IoT devices, allowing users to control functionality, access information or ask questions through voice commands or text messages. It can adapt responses and recommendations based on user preferences, past interactions, and real-time data from sensors or the environment.
Snips.ai: Snips.ai: An open source voice platform that includes an NLU component. It is a prominent platform for developing voice assistants and conversational interfaces for embedded systems and IoT, making it ideal for limited internet connectivity or latency concerns. Its models are optimized for resource-limited embedded devices, requiring minimal RAM, CPU and storage space. This allows integration into most low-power devices. Models can be trained in the language and terminology of a specific domain. The platform prioritizes privacy, local processing and resource efficiency.
Conclusion
Chatbots have many applications in embedded and IoT devices. They play a crucial role in enabling NLP functionalities on edge devices. For simpler tasks, rules-based chatbots can be deployed directly on microcontrollers. However, it may be necessary to transfer NLP functionalities to a cloud-based chatbot service, depending on the complexity of the embedded tasks.
Various machine learning models can also be deployed on microcontrollers using Edge Impulse or Tensor Flow Lite Micro to enable NLP capabilities on microcontroller-based devices. In many applications, the microcontroller is accompanied by a microcomputer inside the device. More complex chatbots can then run on the microcomputer while offloading essential built-in tasks to the microcontroller.
A hybrid approach that combines rules-based chatbots with cloud-based chatbot services is also an option. Some popular chatbots and cloud-based chatbot services include ChatGPT-on-Cloud, Microsoft Azure Bot Service, Dialogflow, ChatScript, AIML, Rasa NLU, ChatterBot, and Snips.ai. Tensor Flow Lite Micro and Edge Impulse are ideal for deploying machine learning models on microcontrollers to implement NLP.
