麻豆传媒

Robot Asimo learns how to jaywalk

New software that tells the humanoid robots where to plant their feet takes inspiration from the classic arcade game, Frogger
Better watch where you step
Better watch where you step
(Image: James Kuffner)

What do the classic computer game Frogger and the movie Raiders of the Lost Ark have in common? They are both being used to demonstrate software that tells humanoid robots where to plant their feet.

At Carnegie Mellon University鈥檚 Robotics Institute in Pittsburgh, Pennsylvania, James Kuffner and his team have developed software that can plan a route through a constantly changing environment. It is used to control an Asimo robot on loan from its Japan-based manufacturer, Honda.

Videos show (4.4mb, .wmv format), and completing a more difficult task in which (45mb, .mov format).

In Frogger, a frog must cross a busy road and avoid getting crushed by cars. In Kuffner鈥檚 life-size, non-lethal version of the game, rows of paper squares moved across the floor by computer-controlled wires take the place of the cars. The software has to find a route that allows the Asimo to reach a coloured disc on the far side of the room without stepping on the moving paper. This is 鈥渢he world鈥檚 first demonstration of autonomous footstep planning for walking humanoids in dynamic environments鈥, says Kuffner.

An overhead camera allows the software to pick out the robot鈥檚 position and the speed and trajectory of each square. The program then constructs a step-by-step map of all the possible routes across the room that are open to the robot. It selects the most advantageous next step for the robot to take to reach its goal smoothly, quickly and avoiding moving obstacles.

So confident is Kuffner of the software that his team is now building something he gleefully calls 鈥渢he Indiana Jones experiment鈥, in which the robot will run a gauntlet of spinning objects and rolling balls somewhat like the hazards the movie character faced. 鈥淲e will be able to make the robot navigate complicated environments that would be difficult even for humans,鈥 he says.

To ready Asimo for this experiment, Kuffner鈥檚 team is improving the vision system and navigation software to allow it to track thin spinning blades from above and calculate their speed. He hopes to run a full version of the experiment this year.

Kuffner鈥檚 software can analyse up to 7000 possible steps per second. It constantly updates its plan as new information is received from the camera, allowing the Asimo to modify its route as the environment changes. The motion of the objects forces the software to consider where they will be by the time the robot reaches them, says Kuffner. 鈥淭his has never been demonstrated on a walking humanoid until now.鈥

鈥淭his has never been demonstrated on a walking robot until now鈥

Honda has previously demonstrated Asimos jogging around an obstacle and climbing stairs, but so far all public exhibitions of the robot have used pre-programmed movements or have been remote-controlled by a human. This would be no use to a robot moving autonomously around a changing environment like a home.

The company does not allow Kuffner to tamper with the Asimo鈥檚 hardware. Instead, an external computer running the route-planning software and connected to the overhead camera controls the robot wirelessly. The team is also developing software that can use a robot鈥檚 on-board head and body cameras.