From “Star Wars” to “Happy Feet,” many beloved films contain scenes that were made possible by motion capture technology, which records movement of objects or people through video. Further, applications for this tracking, which involve complicated interactions between physics, geometry, and perception, extend beyond Hollywood to the military, sports training, medical fields, and computer vision and robotics, allowing engineers to understand and simulate action happening within real-world environments.

As this can be a complex and costly process — often requiring markers placed on objects or people and recording the action sequence — researchers are working to shift the burden to neural networks, which could acquire this data from a simple video and reproduce it in a model. Work in physics simulations and rendering shows promise to make this more widely used, since it can characterize realistic, continuous, dynamic motion from images and transform back and forth between a 2D render and 3D scene in the world. However, to do so, current techniques require precise knowledge of the environmental conditions where the action is taking place, and the choice of renderer, both of which are often unavailable.

Now, a team of researchers from MIT and IBM has developed a trained neural network pipeline that avoids this issue, with the ability to infer the state of the environment and the actions happening, the physical characteristics of the object or person of interest (system), and its control parameters. When tested, the technique can outperform other methods in simulations of four physical systems of rigid and deformable bodies, which illustrate different types of dynamics and interactions, under various environmental conditions. Further, the methodology allows for imitation learning — predicting and reproducing the trajectory of a real-world, flying quadrotor from a video.

“The high-level research problem this paper deals with is how to reconstruct a digital twin from a video of a dynamic system,” says Tao Du PhD ’21, a postdoc in the Department of Electrical Engineering and Computer Science (EECS), a member of Computer Science and Artificial Intelligence Laboratory (CSAIL), and a member of the research team. In order to do this, Du says, “we need to ignore the rendering variances from the video clips and try to grasp of the core information about the dynamic system or the dynamic motion.”

Du’s co-authors include lead author Pingchuan Ma, a graduate student in EECS and a member of CSAIL; Josh Tenenbaum, the Paul E. Newton Career Development Professor of Cognitive Science and Computation in the Department of Brain and Cognitive Sciences and a member of CSAIL; Wojciech Matusik, professor of electrical engineering and computer science and CSAIL member; and MIT-IBM Watson AI Lab principal research staff member Chuang Gan. This work was presented this week the International Conference on Learning Representations.

While capturing videos of characters, robots, or dynamic systems to infer dynamic movement makes this information more accessible, it also brings a new challenge. “The images or videos [and how they are rendered] depend largely on the on the lighting conditions, on the background info, on the texture information, on the material information of your environment, and these are not necessarily measurable in a real-world scenario,” says Du. Without this rendering configuration information or knowledge of which renderer is used, it’s presently difficult to glean dynamic information and predict behavior of the subject of the video. Even if the renderer is known, current neural network approaches still require large sets of training data. However, with their new approach, this can become a moot point. “If you take a video of a leopard running in the morning and in the evening, of course, you’ll get visually different video clips because the lighting conditions are quite different. But what you really care about is the dynamic motion: the joint angles of the leopard — not if they look light or dark,” Du says.

In order to take rendering domains and image differences out of the issue, the team developed a pipeline system containing a neural network, dubbed “rendering invariant state-prediction (RISP)” network. RISP transforms differences in images (pixels) to differences in states of the system — i.e., the environment of action — making their method generalizable and agnostic to rendering configurations. RISP is trained using random rendering parameters and states, which are fed into a differentiable renderer, a type of renderer that measures the sensitivity of pixels with respect to rendering configurations, e.g., lighting or material colors. This generates a set of varied images and video from known ground-truth parameters, which will later allow RISP to reverse that process, predicting the environment state from the input video. The team additionally minimized RISP’s rendering gradients, so that its predictions were less sensitive to changes in rendering configurations, allowing it to learn to forget about visual appearances and focus on learning dynamical states. This is made possible by a differentiable renderer.

The method then uses two similar pipelines, run in parallel. One is for the source domain, with known variables. Here, system parameters and actions are entered into a differentiable simulation. The generated simulation’s states are combined with different rendering configurations into a differentiable renderer to generate images, which are fed into RISP. RISP then outputs predictions about the environmental states. At the same time, a similar target domain pipeline is run with unknown variables. RISP in this pipeline is fed these output images, generating a predicted state. When the predicted states from the source and target domains are compared, a new loss is produced; this difference is used to adjust and optimize some of the parameters in the source domain pipeline. This process can then be iterated on, further reducing the loss between the pipelines.

To determine the success of their method, the team tested it in four simulated systems: a quadrotor (a flying rigid body that doesn’t have any physical contact), a cube (a rigid body that interacts with its environment, like a die), an articulated hand, and a rod (deformable body that can move like a snake). The tasks included estimating the state of a system from an image, identifying the system parameters and action control signals from a video, and discovering the control signals from a target image that direct the system to the desired state. Additionally, they created baselines and an oracle, comparing the novel RISP process in these systems to similar methods that, for example, lack the rendering gradient loss, don’t train a neural network with any loss, or lack the RISP neural network altogether. The team also looked at how the gradient loss impacted the state prediction model’s performance over time. Finally, the researchers deployed their RISP system to infer the motion of a real-world quadrotor, which has complex dynamics, from video. They compared the performance to other techniques that lacked a loss function and used pixel differences, or one that included manual tuning of a renderer’s configuration.

In nearly all of the experiments, the RISP procedure outperformed similar or the state-of-the-art methods available, imitating or reproducing the desired parameters or motion, and proving to be a data-efficient and generalizable competitor to current motion capture approaches.

For this work, the researchers made two important assumptions: that information about the camera is known, such as its position and settings, as well as the geometry and physics governing the object or person that is being tracked. Future work is planned to address this.

“I think the biggest problem we’re solving here is to reconstruct the information in one domain to another, without very expensive equipment,” says Ma. Such an approach should be “useful for [applications such as the] metaverse, which aims to reconstruct the physical world in a virtual environment,” adds Gan. “It is basically an everyday, available solution, that’s neat and simple, to cross domain reconstruction or the inverse dynamics problem,” says Ma.

This research was supported, in part, by the MIT-IBM Watson AI Lab, Nexplore, DARPA Machine Common Sense program, Office of Naval Research (ONR), ONR MURI, and Mitsubishi Electric.

MIT research scientists Pablo Rodriguez-Fernandez and Nathan Howard have just completed one of the most demanding calculations in fusion science — predicting the temperature and density profiles of a magnetically confined plasma via first-principles simulation of plasma turbulence. Solving this problem by brute force is beyond the capabilities of even the most advanced supercomputers. Instead, the researchers used an optimization methodology developed for machine learning to dramatically reduce the CPU time required while maintaining the accuracy of the solution.

Fusion energy

Fusion offers the promise of unlimited, carbon-free energy through the same physical process that powers the sun and the stars. It requires heating the fuel to temperatures above 100 million degrees, well above the point where the electrons are stripped from their atoms, creating a form of matter called plasma. On Earth, researchers use strong magnetic fields to isolate and insulate the hot plasma from ordinary matter. The stronger the magnetic field, the better the quality of the insulation that it provides.

Rodriguez-Fernandez and Howard have focused on predicting the performance expected in the SPARC device, a compact, high-magnetic-field fusion experiment, currently under construction by the MIT spin-out company Commonwealth Fusion Systems (CFS) and researchers from MIT’s Plasma Science and Fusion Center. While the calculation required an extraordinary amount of computer time, over 8 million CPU-hours, what was remarkable was not how much time was used, but how little, given the daunting computational challenge.

The computational challenge of fusion energy

Turbulence, which is the mechanism for most of the heat loss in a confined plasma, is one of the science’s grand challenges and the greatest problem remaining in classical physics. The equations that govern fusion plasmas are well known, but analytic solutions are not possible in the regimes of interest, where nonlinearities are important and solutions encompass an enormous range of spatial and temporal scales. Scientists resort to solving the equations by numerical simulation on computers. It is no accident that fusion researchers have been pioneers in computational physics for the last 50 years.

One of the fundamental problems for researchers is reliably predicting plasma temperature and density given only the magnetic field configuration and the externally applied input power. In confinement devices like SPARC, the external power and the heat input from the fusion process are lost through turbulence in the plasma. The turbulence itself is driven by the difference in the extremely high temperature of the plasma core and the relatively cool temperatures of the plasma edge (merely a few million degrees). Predicting the performance of a self-heated fusion plasma therefore requires a calculation of the power balance between the fusion power input and the losses due to turbulence.

These calculations generally start by assuming plasma temperature and density profiles at a particular location, then computing the heat transported locally by turbulence. However, a useful prediction requires a self-consistent calculation of the profiles across the entire plasma, which includes both the heat input and turbulent losses. Directly solving this problem is beyond the capabilities of any existing computer, so researchers have developed an approach that stitches the profiles together from a series of demanding but tractable local calculations. This method works, but since the heat and particle fluxes depend on multiple parameters, the calculations can be very slow to converge.

However, techniques emerging from the field of machine learning are well suited to optimize just such a calculation. Starting with a set of computationally intensive local calculations run with the full-physics, first-principles CGYRO code (provided by a team from General Atomics led by Jeff Candy) Rodriguez-Fernandez and Howard fit a surrogate mathematical model, which was used to explore and optimize a search within the parameter space. The results of the optimization were compared to the exact calculations at each optimum point, and the system was iterated to a desired level of accuracy. The researchers estimate that the technique reduced the number of runs of the CGYRO code by a factor of four.

New approach increases confidence in predictions

This work, described in a recent publication in the journal Nuclear Fusion, is the highest fidelity calculation ever made of the core of a fusion plasma. It refines and confirms predictions made with less demanding models. Professor Jonathan Citrin, of the Eindhoven University of Technology and leader of the fusion modeling group for DIFFER, the Dutch Institute for Fundamental Energy Research, commented: “The work significantly accelerates our capabilities in more routinely performing ultra-high-fidelity tokamak scenario prediction. This algorithm can help provide the ultimate validation test of machine design or scenario optimization carried out with faster, more reduced modeling, greatly increasing our confidence in the outcomes.” 

In addition to increasing confidence in the fusion performance of the SPARC experiment, this technique provides a roadmap to check and calibrate reduced physics models, which run with a small fraction of the computational power. Such models, cross-checked against the results generated from turbulence simulations, will provide a reliable prediction before each SPARC discharge, helping to guide experimental campaigns and improving the scientific exploitation of the device. It can also be used to tweak and improve even simple data-driven models, which run extremely quickly, allowing researchers to sift through enormous parameter ranges to narrow down possible experiments or possible future machines.

The research was funded by CFS, with computational support from the National Energy Research Scientific Computing Center, a U.S. Department of Energy Office of Science User Facility.