We present a new unsupervised segmentation algorithm, Transition State Clustering (TSC), which combines results from hybrid dynamical systems and Bayesian non-parametric statistics to segment kinematic recordings of robotic surgical procedures.
TSC treats each demonstration trajectory as a noisy observation of an underlying switched linear dynamical system (SLDS) and clusters spatially and temporally similar transition events (i.e., switches in the linear regime). TSC uses a hierarchical Dirichlet Process Gaussian Mixture Model to avoid selecting the number of segments a priori. We compare TSC to five alternatives on the respective algorithms' correspondence to a known ground truth in a synthetic example: Gaussian Mixture Model, Gaussian Hidden Markov Model, Coresets, Gaussian Hidden Semi-Markov Model, and an Autoregressive Hidden Markov Model.
We find that when demonstrations are corrupted with process and observation noise, TSC recovers the ground truth 49% more accurately than alternatives. Furthermore, TSC runs 100x faster than the Autoregressive Models which require expensive MCMC-based inference. We also evaluated TSC on 67 recordings of surgical needle passing and suturing. We supplemented the kinematic recordings with manually annotated visual features denoting grasp and penetration conditions. On this dataset, TSC finds 83% of needle passing transitions and 73% of the suturing transitions annotated by human experts.
- Transition State Clustering: Unsupervised Surgical Trajectory Segmentation For Robot Learning. Sanjay Krishnan, Animesh Garg, Sachin Patil, Colin Lea, Greg Hager, Pieter Abbeel, Ken Goldberg (* denotes equal contribution). Submitted to The International Journal of Robotics Research. May 2016.
- Transition State Clustering: Unsupervised Surgical Trajectory Segmentation For Robot Learning. Sanjay Krishnan, Animesh Garg, Sachin Patil, Colin Lea, Greg Hager, Pieter Abbeel, Ken Goldberg (* denotes equal contribution). International Symposium on Robotics Research (ISRR), 2015.