Menteebot with AI all over its body

Menteebot with AI all over its body

Mentee Robotics states that Menteebot is a personalized artificial intelligence robot that can be guided, with similar balance and control abilities as humans. Just mentioned, unlike previous humanoid robots, Menteebot is covered in AI all over, equipped with rich AI models and algorithms, which allows Menteebot to benefit more from the development of multimodal AI models. Meanwhile, humans can use natural language to control it.
Through the demonstration video, Menteebot's strong AI ability is demonstrated in at least two aspects: first, a more natural communication ability, and second, strong athletic ability.
Looking at communication skills first, Menteebot is equipped with a richer natural language model, which makes its control no longer limited by established instructions. In the past, there was a gap in the ability of humanoid robots to communicate within and outside the model. We have also mentioned this issue before. When a certain instruction falls within the training range of the model, humanoid robots can demonstrate super strong abilities. However, when the instruction exceeds the training range of the model, humanoid robots will appear "at a loss".
It is reported that Menteebot can engage in natural dialogue and communication with humans. Users only need to issue commands to the robot through natural language, and it can understand and execute corresponding tasks. This is actually due to the more powerful artificial intelligence algorithms, large-scale language models, and software in the Menteebot "brain" model. Mentee Robotics states that real-time 3D mapping and positioning based on NeRF, dynamic navigation in complex environments, and other technologies can achieve complex reasoning to complete tasks and quickly learn new tasks.
Mentee Robotics specifically mentioned that deployed Menteebot can undergo more advanced training to cope with complex tasks or scenarios. At this time, the software will continuously simulate the task until it is mastered, and then the robot can complete the task in the real world.
Looking at athletic ability, Menteebot is able to complete more complex walking postures, such as running, lateral walking, and even turning; It can also perform very delicate operations, accurately handing tableware to humans, thanks to its all-round movement ability and precision of arms and hands.
In order to enhance Menteebot's mobility, Mentee Robotics integrates cutting-edge Sim2Real learning on this robot, which can achieve realistic gait and hand movements, with the same balance and control as humans, and can also adjust gait when lifting heavy objects.
Mentee Robotics stated that the mass-produced version of the Menteebot robot is expected to be deployed in the first quarter of 2025, powered by pure visual sensing, specialized electric motors that support "unprecedented" flexibility, and fully integrated artificial intelligence. It is expected to be divided into two versions: the home version and the commercial version.

Menteebot brings inspiration to hardware innovation of humanoid robots

By integrating AI into all operational layers, Menteebot robots demonstrate powerful capabilities, including communication and movement. But this actually puts higher requirements on the deployment of models and algorithms, including the core chips in Menteebot's "brain" and chips in other execution units.
For the core chips in the "brain", the first step is to be able to support multi-modal AI large-scale model deployment, and to leave deployment space for new modalities that will be integrated in the future. At present, the large models applied to humanoid robots include image recognition module, speech recognition module, text to speech module, dialogue system module, navigation module, multimodal system module, and reinforcement learning module. The core chip is the carrier of these modules, and for the above modules, or modes, they need to be able to support them. Even for summary and summarization, the core chip needs to be able to fully support the four major capabilities of visual module, navigation module, language module, and decision module, which puts high requirements on the operator richness of the core chip. After the release of Menteebot, it is expected that reinforcement learning will rise to the fifth fundamental module, and software capabilities will continue to enhance, with hardware redundancy becoming an important indicator.
The second issue is how the execution unit can enhance AI capabilities, which poses new requirements for components such as MCU and FPGA. In terms of MCU, an important development direction for this product category is the ability to deploy and execute AI programs on MCU. In fact, the industry has long been trying to deploy AI on the smallest possible control system, but there are also many challenges in this process.
Firstly, to deploy AI models on the MCU, it is necessary to convert these models into C/C++code, which requires very precise quantification of the models, ensuring their capabilities while avoiding floating-point operations as much as possible. This also has high requirements for MCU compilers, because after quantization, the code needs to be deployed to MCU and needs to go through the compiler.
Secondly, after running the AI model on the MCU, it should not excessively occupy on-chip resources or have too high power consumption. Low power consumption is crucial because there are many MCU devices in a system, and excessive power consumption can affect the overall endurance of the system.
Some people may say that MCU AI is not necessary, but it is very necessary to achieve real-time AI effects.

epilogue

The release of Menteebot can be regarded as the dark horse of the humanoid robot industry, just as stunning as the electric Atlas. Integrating AI into all operational layers is an advanced concept that also requires chips as core hardware to better cope. Among them, high-performance computing chips need to support multimodality, and reinforcement learning modules are expected to rise to core modules; MCU AI is very helpful for improving real-time performance, but it is also very challenging.