Robotics science and systems : online proceedings最新文献

In shared autonomy, user input and robot autonomy are combined to control a robot to achieve a goal. Often, the robot does not know a priori which goal the user wants to achieve, and must both predict the user's intended goal, and assist in achieving that goal. We formulate the problem of shared autonomy as a Partially Observable Markov Decision Process with uncertainty over the user's goal. We utilize maximum entropy inverse optimal control to estimate a distribution over the user's goal based on the history of inputs. Ideally, the robot assists the user by solving for an action which minimizes the expected cost-to-go for the (unknown) goal. As solving the POMDP to select the optimal action is intractable, we use hindsight optimization to approximate the solution. In a user study, we compare our method to a standard predict-then-blend approach. We find that our method enables users to accomplish tasks more quickly while utilizing less input. However, when asked to rate each system, users were mixed in their assessment, citing a tradeoff between maintaining control authority and accomplishing tasks quickly.

Many tasks in robot-assisted surgery, food handling, manufacturing, and other applications require planning and controlling the motions of manipulators or other devices that must interact with highly deformable objects. We present a unified approach for motion planning under uncertainty in deformable environments that maximizes probability of success by accounting for uncertainty in deformation models, noisy sensing, and unpredictable actuation. Unlike prior planners that assume deterministic deformations or treat deformations as a type of small perturbation, our method explicitly considers the uncertainty in large, time-dependent deformations. Our method requires a simulator of deformable objects but places no significant restrictions on the simulator used. We use a sampling-based motion planner in conjunction with the simulator to generate a set of candidate plans based on expected deformations. Our method then uses the simulator and optimal control to numerically estimate time-dependent state distributions based on uncertain parameters (e.g. deformable material properties or actuation errors). We then select the plan with the highest estimated probability of successfully avoiding obstacles and reaching the goal region. Using FEM-based simulation of deformable tissues, we demonstrate the ability of our method to generate high quality plans in two medical-inspired scenarios: (1) guiding bevel-tip steerable needles through slices of deformable tissue around obstacles for minimally invasive biopsies and drug-delivery, and (2) manipulating planar tissues to align interior points at desired coordinates for precision treatment.

Bevel-tip steerable needles are a promising new technology for improving accuracy and accessibility in minimally invasive medical procedures. As yet, 3D needle steering has not been demonstrated in the presence of tissue deformation and uncertainty, despite the application of progressively more sophisticated planning algorithms. This paper presents a feedback controller that steers a needle along 3D helical paths, and varies the helix radius to correct for perturbations. It achieves high accuracy for targets sufficiently far from the needle insertion point; this is counterintuitive because the system is highly under-actuated and not locally controllable. The controller uses a model predictive control framework that chooses a needle twist rate such that the predicted helical trajectory minimizes the distance to the target. Fast branch and bound techniques enable execution at kilohertz rates on a 2GHz PC. We evaluate the controller under a variety of simulated perturbations, including imaging noise, needle deflections, and curvature estimation errors. We also test the controller in a 3D finite element simulator that incorporates deformation in the tissue as well as the needle. In deformable tissue examples, the controller reduced targeting error by up to 88% compared to open-loop execution.