19 December 2011 in The International Journal of Robotics Research
We present parallel algorithms to accelerate collision queries for sample-based motion planning. Our approach is designed for current many-core GPUs and exploits data-parallelism and multi-threaded capabilities. In order to take advantage of the high number of cores, we present a clustering scheme and collision-packet traversal to perform efficient collision queries on multiple configurations simultaneously. Furthermore, we present a hierarchical traversal scheme that performs workload balancing for high parallel efficiency. We have implemented our algorithms on commodity NVIDIA GPUs using CUDA and can perform 500, 000 collision queries per second with our benchmarks, which is 10 times faster than prior GPU-based techniques. Moreover, we can compute collision-free paths for rigid and articulated models in less than 100 ms for many benchmarks, almost 50–100 times faster than current CPU-based PRM planners.
25 May 2010 in Computer Animation and Virtual Worlds
1 November 2013 in ACM Transactions on Graphics
We present a new method for efficiently approximating the global penetration depth between two rigid objects using machine learning techniques. Our approach consists of two phases: offline learning and performing run-time queries. In the learning phase, we precompute an approximation of the contact space of a pair of intersecting objects from a set of samples in the configuration space. We use active and incremental learning algorithms to accelerate the precomputation and improve the accuracy. During the run-time phase, our algorithm performs a nearest-neighbor query based on translational or rotational distance metrics. The run-time query has a small overhead and computes an approximation to global penetration depth in a few milliseconds. We use our algorithm for collision response computations in Box2D or Bullet game physics engines and complex 3D models and observe more than an order of magnitude improvement over prior PD computation techniques.link_to_subscribed_fulltex