Cheap, robust, safe and accepted

By Luca Marchionni, CTO of PAL Robotics

The media has been calling 2024 “the year of the humanoid robots,” and I don’t disagree. The rapid developments in the field are fueled by the boom in artificial intelligence, robotics growth, and a pressing demand for automation across multiple industries. Designed to mimic human movements and perform tasks traditionally handled by people, humanoid robots help address labor shortages and enhance operational versatility.

As the chief technology officer at PAL Robotics, my role is to anticipate the direction of innovation and to build robots that lead the way. Leading our Legged business unit, my focus is to make humanoid robots practical and viable for the market, which expects them to be cheap, robust, safe, and accepted. Keeping up with the competition is part of this role, but it’s becoming more challenging as new humanoids are introduced every few weeks—it’s hard even to keep track of their names.

Versatile or efficient?

Companies like Boston Dynamics, Tesla, Agility Robotics, and Unitree are hinting that bipeds are getting ready for real-world tasks. It has become trendy – and a sign of technological power – to have your own humanoid, and they all need to be smarter, stronger, and more capable than ever. Some of the arguments used to justify this boom are:

• humanoids are versatile and they can solve many of the tasks that aren’t yet automatized
• the human form is the most suitable for a world that is built for humans
• humanoid robots can learn faster and better from human demonstrations than specialized industrial robots

The global market for humanoid is set to grow sharply, from $2.9 billion in 2023 to $46.3 billion by 2031, with an increase of 48.6% per year, according to Coherent Market Insights. It´s unprecedented. But the big question is: will this growth last? Can humanoid robots prove their value in the long run?

Humanoid robots as research platforms

At PAL Robotics we are working towards this goal and we have been commercializing humanoid robots for more than 15 years. Our goal is to provide our customers an ecosystem that allows them to develop and integrate their algorithms and software solutions. We feel proud for having our robots used to train new generations of roboticists, and for that our choice to work with open source tools and libraries such as ROS has a great added value.

Why ROS?

A significant part of our success lies in our commitment to open-source technology. ros_control is the framework we use for real-time control and it originated from the effort of making generic and robot agnostic hardware controllers. We were amongst the first in adopting it and integrating MoveIt and ROS navigation stack on both wheeled and legged platforms.
Our robots’ kinematic and dynamic models, along with simulation tools like Gazebo, MuJoCo, and Isaac Lab, provide a foundation for high-level robotics programming and development.

Our Legged Robots

Most of the complex behaviors of our robots, such as locomotion, manipulation, balance and jumping make use of Whole Body Inverse Dynamics algorithms with joint torque control, which allows them to adapt and be compliant with objects and people. This is an important remark, as humanoid robots are not anchored to the floor: they can slip, collide with obstacles and fall. Interaction forces need to be controlled.

Talos

A torque controlled robot that has been developed in collaboration with LAAS-CNRS and Airbus to explore the use of humanoid robots in the aircraft manufacturing industry. IT is a powerful machine which requirements were:

• EtherCAT communication for fast and precise control of actuators
• Torque sensing for responsive feedback and control
• High payload capacity to handle industrial tools effectively

Kangaroo

Our latest biped robot available as a research platform was designed to be agile and resistant to impacts. Its unique leg architecture places all actuators in the base, using closed kinematic chains to drive the joints. This lightweight design greatly improves power efficiency and enhances stability by minimizing impact forces during ground contact.

Kangaroo’s actuators are equipped with custom force sensors, and its control approach combines Model-Based Control algorithms with cutting-edge Reinforcement Learning (RL) techniques. By training RL algorithms in simulation with GPU acceleration, we’re able to create robust controllers, accelerate development cycles, and push the robot’s hardware to its limits.

Our robots in action

Talos, Kangaroo, and TIAGo Pro were big highlights at the Humanoids 2024 IEEE Conference in Nancy, France. The IEEE-RAS International Conference on Humanoid Robots is an annual event that brings together researchers and industry leaders to discuss advancements in humanoid robotics, covering areas such as mechatronics, control, perception, and human-robot interaction.

This conference offered us a platform to connect with innovators, showcase our work, and exchange insights that will shape the next phase of humanoid development. Check out our wrap up video!