Simulated humanoids present an intriguing platform for investigating motor intelligence with their ability to mimic the full spectrum of human movement. An important area of study in machine learning is the acquisition and application of motor skills. The physical simulation of human talents presents significant control challenges. A controller must deal with a large, unstable, discontinuous system that requires precise timing and coordination to achieve the desired motion.
All current learning methods find it difficult to acquire complicated humanoid behaviors using a tabula rasa approach. Motion capture (MoCap) data has quickly become an integral part of humanoid control studies. MoCap trajectories are sequences of configurations and poses assumed by the human body throughout the movement in question, and as such contain kinematic information about the movement. This information can help a simulated humanoid learn basic motor skills through MoCap demonstrations, making it easier to train complex control strategies.
Unfortunately, use of MoCap data in a physics simulator requires retrieval of the actions (e.g., joint torques) that produce the series of kinematic poses in a given MoCap trajectory (i.e., the trajectory tracking). Finding an action sequence that makes a humanoid track a MoCap sequence is not easy. Reinforcement learning and confrontational learning are two approaches that have been used to address this problem. Training agents to recreate hours of MoCap data is also computationally intensive, and the computational load of detecting these actions grows with the amount of MoCap data. Although MoCap data sets are widely available, few research organizations with considerable computational resources have been able to use them to further learning-based humanoid control.
A recent Microsoft study presented MoCapAct, a high-quality MoCap tracking rule dataset for a MuJoCo-based simulated humanoid, along with a collection of expert implementations of these policies.
With the goal of removing current barriers and enabling the use of MoCap data in humanoid control research, MoCap is designed to be compatible with the very popular dm_control humanoid simulation environment. CMU MoCap is one of the largest publicly available MoCap data sets. MoCapAct policies can track all 3.5 hours of that data.
The researchers demonstrate the use of MoCapAct to learn varied moves by investigating its expert policies and using the expert deployments to train a single hierarchical policy that can track all considered MoCap clips. The low-level part of the policy is then recycled for efficient learning of RL tasks.
The team trained a GPT network to generate a move in the MuJoCo simulator in response to a move prompt using the data set to complete the generative move.
This dataset enables research groups to bypass the time- and energy-consuming process of learning low-level motor skills using MoCap data. This greatly lowers the barrier to entry for simulated humanoid control, promising rich possibilities for exploring multitasking learning and motor intelligence. The team believe their approach can be used in training alternative policy frameworks, such as decision makers, or in settings such as offline reinforcement learning.
This Article is written as a research summary article by Marktechpost Staff based on the research paper 'MoCapAct: A Multi-Task Dataset for Simulated Humanoid Control'. All Credit For This Research Goes To Researchers on This Project. Check out the paper, github, project page and reference article. Please Don't Forget To Join Our ML Subreddit
Tanushree Shenwai is a Consulting Intern at MarktechPost. She is currently pursuing her B.Tech at the Indian Institute of Technology (IIT), Bhubaneswar. She is a data science enthusiast and has a keen interest in the scope of artificial intelligence in various fields. She is passionate about exploring new advances in technologies and their application in real life.
#enable #advanced #research #artificial #humanoid #control #Microsoft #robotics #team #releasing #library #pretrained #simulated #humanoid #control #models #enriched #data #train #robotics