Interactive Motion Correction and Object Manipulation
Ari Shapiro
Marcelo Kallman
University of California, Merced
Petros Faloutsos
We present a system that incorporates a new hybrid motion planning algorithm that: (a) is able to combine keyframe animations with dynamically generated motions, (b) includes time as another dimension in the search space in order to address dynamic environments and achieve synchronization with external motions, and (c) uses efficient configuration sampling strategies for improving the results. We use an analytical inverse kinematics (IK) solution for results an near-interactive rates.
Our method can be applied in two areas: 1) Motion correction - motion capture data is corrected in order to compensate for obstacle in the environment, and 2) objects may be manipulated, grabbed, dropped and touch with high fidelity and respecting constraints such as collisions.

Symposium on Interactive 3D Graphics and Games, 2007
Symposium of Computer Animation, 2006 Posters and Demos
See the entire video
1. Motion Correction
Our motion planner is able to correct motion capture data in order to compensate for objects in the environment. The top image shows the original motion capture walk. An animator has places an umbrella in the hands of the character and added a post. The top images show the character walking normally and the umbrella collides with the post. The bottom image shows the motion after the motion planner has automatically detected collisions for the posts and calculated a new path for the character that respects the virtual objects in the environment.

The animator adds another post to the environment with the same motion capture data. The planner is able to construct a new path for the umbrella that avoids contact with both posts.

Our motion planner searches for collision-free paths around obstacles using an analytical inverse kinematics (IK) solution and Rapidly-exploring Random Trees (RRT). The green path on the first image on the left shows the path of the character's arm as it avoids hitting the vertical post. The other paths in red and brown show the results of various paths being explored during the search. The middle image shows a top view of the same path. Note that the path is choppy and unnatural. Once the path has been calculated, a smoothing process is applied to make the movement simpler and more natural. The last image on the right shows the final path after smoothing. The entire path synthesis process can be performed at near-interactive rates (about 1 second).

Motion can be corrected with moving obstacles and additional characters. The yellow character moves his arms out of the way to avoid the oncoming character. The randomized nature of the planner allows for different ways of obstacle avoidance. The two sequences show two different ways to avoid contact. Biases can be added to force the motion planner to favor one avoidance strategy over the other (for example, "swing the elbow out of the way first")

Legs can be corrected as well as arms. The top image shows an obstacle in the character's path. The botton image shows the corrected motion using our method.

2. Object Manipulation and Grasping
An animator is able to specify grasping and manipulation tasks for characters on top of the original motion capture data. Shown here, the character reaches through an obstacle (the red torus) in order to grab the blue cube. Both the cube and the torus are moving, and the character is able to avoid contact with the torus.

Our system will allow animated character to manipulate objects with simulataneously moving characters, moving obstacles and moving targets. Shown here, the character transitions from a walk to a run while grabbing the blue cube through the red torus while both the torus and the cube are moving. Note that the original motion used is a transition from a walk to a run, while the object manipulation is fully synthesized and attached to the original motion.

The animated character can perform object manipulations with a high degree of accuracy. Here the yellow character steals the hat from the blue character. Notice that the hat is lifted up, avoiding collisions with the head.

Our interactive system allows the animator to specify grabbing and manipulation tasks. Note that the original motion is that of the character moving around the room. The objects were then placed in the room and the character instructed to pick up, put down and reach for various object targets.