The general perception of robotic movement as slow, choppy, and mechanical may soon fall by the wayside as researchers specializing in the digital re-creation of human motion are extending their innovative technology to the robotic realm. Boston Dynamics Inc. (BDI), an R&D company based in Cambridge, Massachusetts, has partnered with Sony Corp. to develop a simulator for modeling the physics and control mechanisms for a prototype robot that Sony is developing for entertainment applications.
The technology, which utilizes real-time 3D graphics to display the simulated behavior, enables Sony engineers to test ideas for robot control and behavior in software, before trying them on the real robot.
The robot simulator is an outgrowth of BDI's human simulation tools. Building on the existing technology required adapting the basic model of human kinematics and geometry to the kinematics and geometry of the Sony robot, according to Marc Raibert, BDI president and former robotics professor at the Massachusetts Institute of Technology, where the BDI technology was born.
"We also had to model the specific electric motors and drive trains used by the robot, rather than what we use for human simulation, and we expanded the ground-contact simulation, so the robot can sit or lie on the ground and behave properly when it gets up."
Critical to the development of the technology has been the use of 3D visualization to analyze the simulation. "When developing a robot or robot-control system, progress can be accelerated by having a good visualization of what is going on," says Raibert. For example, he says, "we sometimes display the simulated forces calculated by the physics as 3D vectors to show where the forces are applied to the body. And it is very useful to vary time when viewing behavior-perhaps moving forward and backward in time to see what happened at a particular event-and coordinating the 3D display with 2D data plots and other quantitative information."
In developing the technology, the BDI engineers grappled with balancing speed and performance. "The challenge was juggling simulator accuracy, run-time performance, and development time. Those three factors trade pretty directly," says Raibert. "You can slow down a simulation in order to get more accurate results, or you can decrease accuracy to get better speed. Or you can do more engineering to improve both accuracy and speed." On this project, he says, "we made it so the end users can adjust the trade-offs themselves to accommodate the situation at hand."
|Walk this way. A prototype Sony robot is taking steps toward commercialization thanks to a physics-based simulation system that gives it human-like motion.|
A more significant obstacle, Raibert believes, is one that is universal to robotic simulation on the whole. "One of the challenges of these simulators is to create control systems that produce human-like behavior. We don't want 'corpse' simulators, we want simulators that create lifelike motion."
As an ongoing research project, BDI's robot simulator is constantly evolving. On the agenda are the integration of additional behavioral capabilities, interface enhancements, and interactive visualization capabilities that will let users provide input to the simulated behavior to attain more lifelike motion and automatically modify the simulation appropriately.
Because the Sony Robot is not yet a product, the commercial future of the simulator is hard to predict. "I do think people who buy robots will want simulators too," says Raibert. And Sony's focus on the entertainment sector will likely make the demand for robots more palpable. "This application is more 'real' than factory automation in terms of potential market, interesting robot behavior, and technology." The potential exists for entertainment robots to be the vehicle that carries high-tech robots to the masses, Raibert notes. "Wouldn't it be cool if physics-based robot simulation went there too?"
Diana Phillips Mahoney is chief technology editor of Computer Graphics World.
|A graphical interface enables users to analyze and direct the robot's behavior during a simulation.|