Basketball is one of the most popular sports around. And it has inspired many cartoons. The cartoons are fun to watch and encourage children to play the sport.
The research team at Carnegie Mellon University is working on a program that lets animated basketball players practice dribbling. These avatars learn through millions of practice repetitions.
Physically Simulated Dribbling Skills
Despite advances in graphics and character animation in games like EA’s NBA LIVE and NBA 2K, basketball video games still rely heavily on canned animations. The industry is always looking for new methods to create gripping, on-court action in a more personalized, interactive way.
But getting the small details right — such as how the ball may spin when a player’s fingers make light contact — isn’t always easy. That’s especially true for dribbling the basketball.
Researchers at DeepMotion and Carnegie Mellon University have developed a method to enable animated characters in video games to learn their basketball dribbling skills from motion capture moments of real people performing dribbling exercises. This trial-and-error learning process requires millions of trials, but results in arm movements that are closely coordinated with physically plausible ball movement.
The physics-based approach to learning isn’t perfect, though, since it relies on a complex set of algorithms that don’t work for all sports and motions. The teams are currently working on a system that uses trajectory optimization to estimate where the ball is most likely to move for each hand movement.
They’re also working on a method to train players how to transition from one skill to another without losing the ball. These skills, called transitions, are critical to basketball play and can make or break a game.
To train these transitions, researchers use biomechanical modeling to ascribe joints and torque to the character rig (Deep Motion’s Avatar Physics Engine uses similar “articulated physics”). Different joint sets can be trained separately for targeted skill development.
Once the character is learned how to perform these specific motions, they can then be taught more advanced techniques. This can include how to dribble between their legs, behind their backs and in crossover moves. It can also teach them how to transition from one dribbling skill to the next without losing the ball.
In addition to dribbling, basketball players are also expected to perform high-tempo sprints. This is an important skill to have because it’s a critical component of fast breaks, or transitions from defense to offense.
Multi-Skill Control Graph
Basketball is a dynamic sport that requires a range of skills. A great player can move fluidly between dribbling skills, using quick pivoting and fake outs. This can be challenging for physics-based animation, especially in competitive environments.
To create a character with a variety of ball handling skills, researchers have developed a method to train a multi-skill control graph. This method allows the user to transition between motion fragments of different skills, enabling a smooth transition from one skill to another without causing glitches. This is a significant improvement over existing methods which only allow for motion re-compositions of single dribbling skills.
The multi-skill control graph is based on a combination of joint set training and biomechanical modeling. By assigning each simulated joint torque and force to the corresponding muscle, the character’s body state can be physically modeled. The resulting model is then used to control the character’s movements.
Liu and Hodgins demonstrate that the resulting multi-skill control graph can be trained by incrementally training different skill sets, thereby ensuring that the character is capable of seamlessly blending skills. This technique is called “articulated physics” and was first developed by Deep Motion.
While articulated physics can be trained for the entire character, the authors also showed that it can be effectively used for training individual skills such as dribbling and shooting. These skills can then be adapted to the environment in which the characters will be playing.
For example, the authors trained a dribbling skill that uses an arc to hit the defender and then pivots away from the defender. This method was then applied to an eSports game. The results were impressive and allowed the simulated character to perform a number of dribbling skills including the ability to shoot in a variety of different ways.
This technique has also been applied to football and tennis. The researchers have found that the resulting dribbling controllers are robust and can be applied to multiple games.
Each part of a basketball team is engineered to perform a specific task, and different players are gifted with different abilities. This is the principle behind the design and development of animated basketball players, which can be used as teaching and training resources. The animated characters are designed to simulate the movement process of key bones in basketball athletes, and the system can automatically control their training according to the motion information captured by sensors.