Alla Safonova

Achieving good connectivity in motion graphs

Motion graphs provide users with a simple yet powerful way to synthesize human motions. While motion graph-based synthesis has been widely successful, the quality of the generated motion depends largely on the connectivity of the graph and the quality of transitions in it. However, achieving both of these criteria simultaneously in motion graphs is difficult. Good connectivity requires transitions between less similar poses, while good motion quality results only when transitions happen between very similar poses. This paper introduces a new method for building motion graphs. The method first builds a set of interpolated motion clips, which contain many more similar poses than the original dataset. Using this set, the method then constructs a motion graph and reduces its size by minimizing the number of interpolated poses present in the graph. The outcome of the algorithm is a motion graph called a well-connected motion graph with very good connectivity and only smooth transitions. Our experimental results show that well-connected motion graphs outperform standard motion graphs across a number of measures, result in very good motion quality, allow for high responsiveness when used for interactive control, and even do not require post-processing of the synthesized motions.

Human Motion Synthesis with Optimization‐based Graphs

Continuous constrained optimization is a powerful tool for synthesizing novel human motion segments that are short. Graph‐based motion synthesis methods such as motion graphs and move trees are popular ways to synthesize long motions by playing back a sequence of existing motion segments. However, motion graphs only support transitions between similar frames, and move trees only support transitions between the end of one motion segment and the start of another. In this paper, we introduce an optimization‐based graph that combines continuous constrained optimization with graph‐based motion synthesis. The constrained optimization is used to create a vast number of complex realistic‐looking transitions in the graph. The graph can then be used to synthesize long motions with non‐trivial transitions that for example allow the character to switch its behavior abruptly while retaining motion naturalness. We also propose to build this graph semi‐autonomously by requiring a user to classify generated transitions as acceptable or not and explicitly minimizing the amount of required classifications. This process guarantees the quality consistency of the optimization‐based graph at the cost of limited user involvement.

High-dimensional planning on the GPU

Optimal heuristic searches such as A* search are commonly used for low-dimensional planning such as 2D path finding. These algorithms however, typically do not scale well to high-dimensional planning problems such as motion planning for robotic arms, computing motion trajectories for non-holonomic robotic vehicles and motion synthesis for humanoid characters. A recently developed randomized version of A* search, called R* search, scales to higher-dimensional planning problems by trading off deterministic optimality guarantees of A* for probabilistic sub-optimality guarantees. In this paper, we show that in addition to its scalability, R* lends itself well to a parallel implementation. In particular, we demonstrate how R* can be implemented on the GPU. On the theoretical side, the GPU version of R*, called R*GPU, preserves all the theoretical properties of R* including its probabilistic bounds on sub-optimality. On the experimental side, we show that R*GPU consistently produces lower cost solutions, scales better in terms of memory, and runs faster than R*. These results hold for both motion planning for a 6DOF robot arm planar as well as 2D path finding.

Automatic construction of a minimum size motion graph

Motion capture data have been used effectively in many areas of human motion synthesis. Among those, motion graph-based approaches have shown great promise for novice users due to their ability to generate long motions and the fully automatic process of motion synthesis. The performance of motion graph based approaches, however, relies heavily on selecting a good set of motions used to build the graph. This motion set needs to contain enough motions to achieve good connectivity and smooth transitions. At the same time, the motion set needs to be small for fast motion synthesis. Manually selecting a good motion set that achieves these requirements is difficult, especially given that motion capture databases are growing larger to provide a richer variety of human motions. Therefore we propose an automatic approach to select a good motion set. We cast the motion selection problem as a search for a minimum size subgraph from a large motion graph representing the motion capture database and propose an efficient algorithm, called the Iterative Sub-graph Algorithm, to find a good approximation to the optimal solution. Our approach especially benefits novice users who desire simple and fully automatic motion synthesis tools, such as motion graph-based techniques.

Planning with adaptive dimensionality for mobile manipulation

Mobile manipulation planning is a hard problem composed of multiple challenging sub-problems, some of which require searching through large high-dimensional state-spaces. The focus of this work is on computing a trajectory to safely maneuver an object through an environment, given the start and goal configurations. In this work we present a heuristic search-based deterministic mobile manipulation planner, based on our recently-developed algorithm for planning with adaptive dimensionality. Our planner demonstrates reasonable performance, while also providing strong guarantees on completeness and suboptimality bounds with respect to the graph representing the problem.

Synthesizing Human Motion from Intuitive Constraints

Many compelling applications would become feasible if novice users had the ability to synthesize high quality human motion based only on a simple sketch and a few easily specified constraints. Motion graphs and their variations have proven to be a powerful tool for synthesizing human motion when only a rough sketch is given. Motion graphs are simple to implement, and the synthesis can be fully automatic. When unrolled into the environment, motion graphs, however, grow drastically in size. The major challenge is then searching these large graphs for motions that satisfy user constraints. A number of sub-optimal algorithms that do not provide guarantees on the optimality of the solution have been proposed. In this paper, we argue that in many situations to get natural results an optimal or nearly-optimal search is required. We show how to use the well-known A* search to find solutions that are optimal or of bounded sub-optimality. We achieve this goal for large motion graphs by performing a lossless compression of the motion graph and implementing a heuristic function that significantly accelerates the search for the domain of human motion. We demonstrate the power of this approach by synthesizing optimal or near optimal motions that include a variety of behaviors in a single motion. These experiments show that motions become more natural as the optimality improves.

GPU methods for real-time haptic interaction with 3D fluids

Real-time haptic rendering of three-dimensional fluid flow will improve the interactivity and realism of applications ranging from video games to surgical simulators, but it remains a challenging undertaking due to its high computational cost. Humans are very familiar with the look and feel of real fluids, so successful interactive simulations need to obey the mathematical relationships of fluid dynamics with high spatial resolution and fast temporal response. In this work we propose an innovative GPU-based approach that enables real-time haptic rendering of high-resolution 3D Navier-Stokes fluids. We show that moving the vast majority of the computation to the GPU allows for the simulation of touchable fluids at resolutions and frame rates that are significantly higher than any other recent real-time methods without a need for pre-computations. Based on our proposed approach, we build a haptic and graphic rendering system that allows users to interact with 3D virtual smoke in real time through the Novint Falcon, a commercial haptic interface.

Stochastic activity authoring with direct user control

Crowd activities are often randomized to create the appearance of heterogeneity. However, the parameters that control randomization are frequently hard to tune because it is unclear how changes at the character level affect the high-level appearance of the crowd. We propose a method for computing randomization parameters that supports direct animator control. Given details about the environment, available activities, timing information and the desired high-level appearance of the crowd, we model the problem as a graph, formulate a convex optimization problem, and solve for a set of stochastic transition rates which satisfy the constraints. Unlike the use of heuristics for adding randomness to crowd activities, our approach provides guarantees on convergence to the desired result, allows for decentralized simulation, and supports a variety of constraints. In addition, because the rates can be pre-computed, no additional runtime processing is needed during simulation.

A data-driven appearance model for human fatigue

Humans become visibly tired during physical activity. After a set of squats, jumping jacks or walking up a flight of stairs, individuals start to pant, sweat, loose their balance, and flush. Simulating these physiological changes due to exertion and exhaustion on an animated character greatly enhances a motions realism. These fatigue factors depend on the mechanical, physical, and biochemical function states of the human body. The difficulty of simulating fatigue for character animation is due in part to the complex anatomy of the human body. We present a multi-modal capturing technique for acquiring synchronized biosignal data and motion capture data to enhance character animation. The fatigue model utilizes an anatomically derived model of the human body that includes a torso, organs, face, and rigged body. This model is then driven by biosignal output. Our animations show the wide range of exhaustion behaviors synthesized from real biological data output. We demonstrate the fatigue model by augmenting standard motion capture with exhaustion effects to produce more realistic appearance changes during three exercise examples. We compare the fatigue model with both simple procedural methods and a dense marker set data capture of exercise motions.

Motion planning for robotic manipulators with independent wrist joints

Advanced modern humanoid robots often have complex manipulators with a large number of degrees of freedom. Thus, motion planning for such manipulators is a very computationally challenging problem. However, often robotic manipulators allow the wrist degrees of freedom to be controlled independently from the configuration of the rest of the arm. In this paper we show how to split the high dimensional planning problem into two lower-dimensional sub-problems - planning for the main arm joints and planning for the wrist joints, without losing guarantees on completeness. This approach is an extension of our previously developed framework for planning with adaptive dimensionality. Experimentally, we show that this approach is very effective in speeding up planning for robotic arms on Willow Garages PR2 platform. We compare our algorithm with several popular alternative approaches for performing motion planning for robotic arms. The results we observe illustrate that our algorithm provides a good balance between planning time, planning success rate, path consistency, and path quality.