CHANNELS

Aerospace

Automotive

Automated & Autonomous Vehicles

Battery & Electrification Technology

Defense

Drone TV

Electronics

Manufacturing & Prototyping

Materials

Medical

Motion Control

RF & Microwave Technology

Robotics & Automation

Sensors & IoT

Test & Measurement

Aerospace
Automotive
Automated & Autonomous Vehicles
Battery & Electrification Technology
Defense
Drone TV
Electronics
Manufacturing & Prototyping
Materials
Medical
Motion Control
RF & Microwave Technology
Robotics & Automation
Sensors & IoT
Test & Measurement
Aerospace & Defense
Search Results

Speedy Processor for "Motion Planning" Helps Robots Navigate Obstacles

At toddler age, humans are pretty good at what roboticists call 'motion planning' - reaching around obstacles to precisely pick up objects both seen and unseen. But for robots with multi-jointed arms, motion planning is a difficult problem that requires time-consuming computation. Picking an object up in an environment that has not been pre-engineered for the robot may require several seconds of computation. Duke University researchers introduce a specially-designed computer processor for motion planning that can plan up to 10,000 times faster than existing approaches while consuming a small fraction of the power. The new processor is fast enough to plan and operate in real time, and power-efficient enough to be used in large-scale manufacturing environments with thousands of robots. Speedy motion planning saves the time and expense of engineering the environment around the robot. The team designed their new processor to perform collision detection - the most time-consuming aspect of motion planning - such that the processor performs thousands of collision checks in parallel. The new processor's speed and power efficiency could create many opportunities for automation.

Topics:
Automation Electronics & Computers Manufacturing & Prototyping Motion Control Robotics Software

More From SAE Media Group

    Transcript

    00:00:01 motion planning is the problem of finding a path to move a robot from its current position to some goal position without colliding with any obstacles here we see a robotic arm moving around a challenging set of obstacles to reach out and grasp the pink toy the ability to do this quickly is critical for robots that must operate in environments like the home that are not carefully

    00:00:21 controlled and structured although motion planning has been studied for decades existing techniques take seconds on general purpose CPU and hundreds of milliseconds on power hungry gpus to create a single plan this is a major obstacle to the Practical use of robots in unstructured environments as there are often long delays between

    00:00:42 the desire to execute a motion represented here by pressing the red button and actually having a plan to execute most of this time is spent doing thousands or millions of calculations checking many small movements to see whether or not they collide with any of the obstacles present our team has developed a specialized motion planning processor

    00:01:03 that is able to find motion plans in under a millisecond motion planning is usually done by building a road map which is just a network where each node represents a specific position of the robot for example this pose with the arm outstretched above the table is represented by the red node in the network the green node represents a different pose with the hand placed

    00:01:25 directly over the sphere two nodes in the network can be connected if the robot can move between the the two positions safely this Edge represents a motion between the two poses seen here an actual road map may have many thousands of nodes and edges as it moves the robot sweeps a volume through space as can be seen by visualizing many intermediate poses of the robot to avoid

    00:01:50 performing expensive Collision checking at runtime we discretize space into voxal and for each Motion in the road map we compute the set of voxal that motion collides with the set for the motion shown earlier is visualized here this set of voxal represents the volume swept by the arm we use these sets to build specialized circuits to detect collisions obstacle data is streamed

    00:02:14 over the circuit which simply output a true or false value of whether that motion is in Collision we build such a circuit for each Edge in the road map at runtime when a robot perceives the environment all the Motions can be checked for collisions simultaneously in parallel we Implement these circuits on an fpga for use with the robotic arm shown

    00:02:36 earlier perception of the environment is done using these four Microsoft connects from which we can determine which obstacle voxal are present the Collision detection doesn't begin until the red button is pressed you can see the arm begins moving immediately with no delay our technology is up to 10,000 times faster than previous techniques

    00:03:02 and consumes 15 times less energy the new approach is so fast that motion planning can now be done in real time faster than perception and used as a sub routine to perform more complex tasks opening up many opportunities for automation