GetRobo Blog English

New Artificial Muscles to Generate Human Movements in Robots

  One of the reasons you realize that a robot like this is indeed artificial and not a human being is because it's movements are jumpy and jerky. So what exactly is the difference between a mechanical movement and a "natural" one?  And is it possible to make a robot move more "naturally?"

   Jun Ueda, Assistant Professor at the Georgia Institute of Technology (photo), is set to find that out.
 
  In a human being, every time there is an electric signal from the brain - let's say - to wave an arm, different cells respond to that signal. The stochastic - or random - way the cells respond to a certain signal explains why every wave of the arm is slightly different from the last, which is something that doesn't occur with robots. A robot's wave will be exactly the same every time.
 
  So Prof. Ueda is developing a "stochastically-controlled, biologically-inspired actuator" to see if he can imitate the stochastic human process in robots. With current artificial muscles, if you send a signal at a regular interval, the movement will be jerky. Prof. Ueda thinks with his new cell-structured actuator combined with the right timing of electric pulses, he could generate a natural motion artificially.

(Photo: Actuator using 6 piezoelectric devices.) 
 
  The new actuator is made from an array of small piezoelectric devices (cells) that are linked together to build a muscle-like formation. Each device can only move a tiny bit depending on the electric voltage but linking many of them together and covering them with a mechanical frame enables it to make bigger strokes as a whole. The new actuator can respond more quickly than the existing artificial muscles which use air pumps or shape memory alloys. The cell structure is also useful in a way that you can change the length and size of the artificial muscle, as well as fix it more easily by replacing the tiny devices.

(Photo: Goal is to create a human muscle-like structure using a number of the small actuators, as seen in the model on the left. Sending electric signals to each actuator (cell) randomly may realize movements that are closer to humans.)
   
  Besides the intellectual merit of finding out the "missing link" that bridge the gap between biological movements and artificial ones, Prof. Ueda is eager to develop a robotic arm driven by a number of stochastic array actuators.

Humanoid HRP-2 Walks on Uneven Terrain

 Many people used to think that two-legged humanoid robots with motors in their joints would never be able to walk on uneven surfaces unless they knew exactly in advance what they were stepping on and when. Drop a small pebble in its path and you could topple the robot over. Unlike in the movies, bipedal walking for robots in real life is extremely difficult.
 
 Dr. Koichi Nishiwaki and his colleagues at the AIST Digital Human Research Center have been working hard to dispel this accepted myth. GetRobo got to see a demo of the humanoid robot HRP-2  walking on pebbles, planks and tiles, at the DHRC in Tokyo.

 
 In this video, Dr. Nishiwaki is telling the robot where to go using a joystick, but other than that, the HRP-2 is on it's own. The robot has what is called an "attitude measurement sensor" on it's waist which is used to measure and control the posture as well as to estimate the absolute motion. Depending on the feedback from this sensor, a computer inside the robot generates a dynamically stable motion pattern 50 times per second. The robot checks and adjusts to this desirable motion 1,000 times per second, enabling it to remain stable and maintain it's walk.
 
 To actually reproduce the desirable motion, instead of strictly trying to control where and how the robot's foot should be placed, Dr. Nishiwaki uses a method in which the force the foot receives from the floor is checked and managed.

 

 Dr. Nishiwaki was able to make the robot remain stable even when the robot stepped on the edge of a plank (as seen in the beginning of the video) - a feat that no one else had been able to accomplish before.
 
 Still there is more to overcome before robots can really walk around your house and neighborhood. (For example, this robot can't cross it's legs to maintain a balance in certain situations like a human would.)  So when is that going to happen? 

 

 " Combined with environment recognition and path planning technologies, in 3 to 5 years, I would like to realize a humanoid robot that can walk freely indoors and on paved roads," says Dr. Nishiwaki. Now THAT'S not too far away.

 

(Photo: Dr. Nishiwaki = left = with HRP-2 and post-doctoral researcher Joel Chestnutt.

Next we would like to see HRP-2 walk on THOSE steps, don't we? Nishiwaki-san, yoroshiku onegai shimasu!)

Bots Compete in Rescue Competition

 The finals of the 10th Rescue Robot Contest is happening in Kobe, Japan this weekend. Twelve teams and their robots that made through the preliminaries on July 4th will be competing on time and skill to "rescue" silicon rubber dolls from rubbles.

 GetRobo was fortunate enough to be in Kobe to see the preliminaries and wanted to share some photos.  

 The competition field is a 1/6 size mock-up of a disaster stricken area. (As many will recall, Kobe was hit by the Great Hanshin Earthquake in 1995 which claimed more than 6,000 people. The contest initiated to raise awareness of disaster prevention and the hope that robotics will be of help in future undesirable events.) At the preliminaries, one doll was laid under wooden pillars and another was situated inside a house. Each team was given 8 minutes for the robots to remove the rubbles, roof and/or walls of the house and then carry the dolls safely to the goal without causing too much negative impact onto their bodies.

 

(Photo1: A robot by Team Kobe City College of Technology pulls out the injured.)

 Since there are no restrictions on how many robots you can use, all teams had multiple robots which had different roles.

 

(Photo2: Two robots from Team Rescue Gorilla B of Osaka Electro-Communication University each carrying a doll)

 

(Photo3: Team Rescue Gorilla B's 3rd robot laid out a carpet over the bumps so that it can reduce vibration which may have a negative impact on the dolls while being carried by the other 2 robots.)

 The robots are remotely controlled using WiFi. Team members were allowed to look directly at the field to control the robots during the preliminaries.

 

(Photo4: Team Rescue Gorilla S, also of OECU tries to figure out how to rescue the doll inside the house.)

 Although, the finals will be more difficult because teams are only allowed to use the visions of the cameras on the robots and one set up from above by the contest organizer. So some teams practiced trying to set up an extra camera in the field to obtain additional visual data.