Adam, a state-of-the-art humanoid robot, is setting new standards in robotic mobility with a cutting-edge reinforcement learning (RL) algorithm. Developed by PNDbotics, Adam is designed to master human-like locomotion through an innovative algorithm refined via extensive simulations and vast training data.
Since its inception in June 2023, the PNDbotics team has been tirelessly working on enhancing Adam’s core components, particularly focusing on strengthening its legs and feet for durability across varied environments. The modular architecture of its actuators significantly boosts Adam’s flexibility, allowing it to better adapt to dynamic and ever-changing conditions.
The PNDbotics website highlights their adoption of the latest Deep Reinforcement Learning (DRL) and imitation learning algorithms. These advanced techniques provide developers with easy access to the NVIDIA Isaac Gym parallel DRL training environment, enabling the development of individual algorithms tailored to specific needs.
Adaptive Robot Evolution
Traditional robot control algorithms, which rely on precise mathematical models and predefined motion planning, have proven effective in locomotion. Notable examples include Boston Dynamics’ Atlas and Spot robots using Model Predictive Control (MPC). However, these methods often face challenges in unknown or dynamically changing environments, where reliance on accurate modeling limits adaptability and complicates development.
To address these challenges, researchers are increasingly turning to deep RL, which offers robots greater flexibility and resilience. This approach allows robots to learn strategies through interaction with their surroundings independently. While RL has shown great promise in legged robots, its application in humanoid robots has been limited by high costs, maintenance issues, and the difficulty of applying learned models to real-world scenarios.
The PNDbotic team developed Adam, a motor-joint-driven humanoid robot with human-like motion, high-performance actuators, and cost-effective modularity, to tackle these issues. Adam employs an imitation learning framework and human motion data to achieve remarkable locomotion performance. This development not only makes humanoid robotics research more accessible but also opens new avenues for learning, adapting, and optimizing in challenging settings.
Reinventing Robot Mobility
Adam (Lite) stands at 1.6 meters tall and weighs 60 kilograms. It boasts 25 quasi-direct drive (QDD) force-controlled PND actuators, which provide exceptional mobility and adaptability. Its legs are equipped with high-sensitivity actuators capable of up to 360 Nm torque, while the arms offer five degrees of freedom and the waist three.
Adam’s modular design, biomimetic torso, and a specially designed full-stack system, which includes an Intel i7-powered PND Robot Control Unit and a real-time PDN network, facilitate advanced full-body motion control. This configuration prepares Adam for real-world service scenarios, easing neural network training and extensive dynamic simulations. Although vision modules and dexterous hands can be added, the primary focus of this project has been on blind movement.
To enhance training, the team meticulously calibrated and converted data to fit Adam’s specific structure, adding high-precision motion capture of specially created movements to public datasets. This approach ensured adaptability across various complex tasks by enabling precise model optimization.
Enhanced Adaptability and Efficiency
Adam’s RL-driven approach provides remarkable adaptability, setting it apart from traditional model-based gait control. It can adjust its stride, pace, and balance in real time on challenging terrains like uneven surfaces, ensuring stable and natural movement.
According to PNDbotics, Adam’s adaptability enhances both efficiency and safety. The RL algorithm also facilitates fluid motion with smooth joint coordination, reducing mechanical wear, extending operational life, and improving energy efficiency for enhanced endurance.
Note: This article is inspired by content from https://interestingengineering.com/innovation/chinas-humanoid-robot-walks-like-human. It has been rephrased for originality. Images are credited to the original source.