Guiding needle insertions using tissue manipulation

This project investigates the use of robotic tissue manipulation in medical needle insertion procedures, in order to improve targeting accuracy and help avoid damaging sensitive tissues. Our approach integrates imaging feedback and finite element tissue modeling to control points of interest in soft tissue. Automated procedure planning is used to select optimal manipulation devices and placement locations. Our simulations of prostate brachytherapy suggest that tissue manipulators can correct fot disturbances caused by needle placement errors and tissue deflections, and may greatly reduce trauma caused by needle puncture of the urethra and seminal vesicles.

Control of steerable needles in deformable tissue

Bevel-tip steerable needles are a promising new technology for improving accuracy and accessibility in minimally invasive medical procedures. In collaboration with researchers at UC Berkeley, UNC Chapel Hill, and Johns Hopkins, we are developing motion planning and feedback control algorithms for 3D needle steering in the presence of tissue deformation and uncertainty. We evaluate our algorithms in state-of-the-art 3D finite element simulations that incorporates deformation in the tissue as well as the needle. Compared with open-loop execution, our feedback strategies reduce targeting error by up to 88%.

Motion planning for legged robots on rough terrain

In collaboration with researchers at JPL, AIST Japan, and Stanford, we study the quasi-static motion of large legged robots that have many degrees of freedom. While gaited walking may suffice on easy ground, rough terrain requires unique sequences of footsteps and postural adjustments adapted to the terrain's local geometric and physical properties. We are developing planners that compute these motions by combining graph searching to generate a sequence of candidate footfalls with probabilistic sample-based planning to generate continuous motions that reach these footfalls. We have assessed the viability of this approach in simu- lation for the six-legged lunar vehicle ATHLETE and the humanoid HRP-2, as well as experiments on the Capuchin rock climbing robot.

Planning using high-quality motion primitives

We are investigating methods of computing efficient and natural-looking motions for humanoid robots walking on varied terrain. We consider the use of a small set of high-quality motion primitives (such as a fixed gait on flat ground) that have been generated offline. But rather than restrict motion to these primitives, it uses them to derive a sampling strategy for a probabilistic, sample-based planner. Results in simulation on several different terrains demonstrate a reduction in planning time and a marked increase in motion quality.

Leaving Flatland: Real-time planning and system integration in 3D environments

In collaboration with researchers at SRI and Boston Dynamics, the Leaving Flatland project attempts to surmount the challenges of closing the loop between autonomous perception and action on challenging terrain. The proposed system includes comprehensive localization, mapping, path planning and visualization techniques for a mobile robot to operate autonomously in complex 3D indoor and outdoor environments. In doing so we integrate robust Visual Odometry localization techniques with real-time 3D mapping methods from stereo data to obtain consistent global models annotated with semantic labels. These models are used by a multi-region motion planner which adapts existing 2D planning techniques to operate in 3D terrain.

Object pushing on a humanoid robot

In collaboration with Honda, we are investigating algorithms that enable the Honda ASIMO humanoid robot to push an object to a desired location on a cluttered table. ASIMO's existing hardware configuration was not designed specifically for manipulation: its arms are short and have only 5 degrees of freedom, and its cameras cannot look downward. Thus, visual sensing can only be performed infrequently, and the robot must frequently walk to reposition itself between pushes. We have developed a motion planner that uses a restricted class of stable pushes, and can solve problems that require several carefully chosen pushes in minutes. Our algorithm has been evaluated in simulation and on the Honda ASIMO robot. Current work addresses multi-handed pushes, imprecise pushes, and integration with other modes of manipulation.

Completeness of multi-modal motion planning

The motion planning problems encountered in manipulation and legged locomotion have a distinctive multi-modal structure, where the space of feasible configurations consists of overlapping submanifolds of differing dimensionality. Previously developed planners have questionable reliability. They may rely on incomplete heuristics to select a good sequence of submanifolds to explore, or may require setting parameters that trade off speed for how accurately the feasible space is represented. We are developing new multi-modal planning algorithms with strong theoretical completeness properties. Such algorithms may enable reliable execution of difficult tasks, e.g. automated assembly, household object manipulation, and surgical manipulation.