Search

When searching for objects in cluttered environments, it is often necessary to perform complex interactions in order to move occluding objects out of the way and fully reveal the object of interest and make it graspable. Due to the complexity of the physics involved and the lack of accurate models of the clutter, planning and controlling precise predefined interactions with accurate outcome is extremely hard, when not impossible. In problems where accurate (forward) models are lacking, Deep Reinforcement Learning (RL) has shown to be a viable solution to map observations (e.g. images) to good interactions in the form of close-loop visuomotor policies. However, Deep RL is sample inefficient and fails when applied directly to the problem of unoccluding objects based on images. In this work we present a novel Deep RL procedure that combines i) teacher-aided exploration, ii) a critic with privileged information, and iii) mid-level representations, resulting in sample efficient and effective learning for the problem of uncovering a target object occluded by a heap of unknown objects. Our experiments show that our approach trains faster and converges to more efficient uncovering solutions than baselines and ablations, and that our uncovering policies lead to an average improvement in the graspability of the target object, facilitating downstream retrieval applications.

Introduction: The posterior ventricular anchoring neochordal (PVAN) repair is a nonresectional, single-suture technique for correcting posterior leaflet prolapse. While this technique has demonstrated clinical efficacy, a possible limitation is the stability of the suture anchored into myocardium as opposed to the fibrous portion of a papillary muscle. Hypothesis: We hypothesize that the PVAN suture serves only to position the leaflet for coaptation, after which systolic forces will be distributed throughout the valve, resulting in low peak forces on the suture. Methods: A left heart simulator was constructed using 3D printing, tuned to generate physiological pressure and flow waveforms, then validated. Porcine mitral valves (n=9) were dissected and mounted within the simulator. Chordal forces were measured using Fiber Bragg Grating (FBG) sensors, sewn in place using PTFE suture. FBG sensors are strain gauges made of 125 μ m optical fibers that use reflected peak wavelength changes to measure strain. Hemodynamic and echocardiographic data were also collected. Isolated severe mitral regurgitation (MR) was induced by cutting P2 primary chordae. The valve was repaired using the PVAN technique, anchoring the suture to a customized force-sensing post positioned to mimic in vivo placement. Results: Forces on 1° and 2° chordae of both anterior and posterior leaflets were significantly elevated in the prolapse condition ( P < 0.05). PVAN resulted in elimination of MR in all valves, as well as normalization of chordae forces to baseline levels for posterior primary ( P < 0.01 ) , posterior secondary ( P < 0.01 ) , and anterior primary chordae ( P < 0.05 ) , with reduction in anterior secondary chordal forces approaching significance ( P = 0.055 ) . Peak forces on the PVAN stitch were minimal, even compared to the forces experienced by primary chordae of normal, healthy valves ( P < 0.05). Conclusions: The PVAN technique eliminates MR by effectively positioning the posterior leaflet for optimal coaptation, distributing the forces amongst the subvalvular apparatus. Given the extremely low forces involved, the strength of the ventricular anchoring suture and myocardial anchoring point should not be a limiting factor.

Introduction: Very few technologies exist that can provide quantitative data on forces within the mitral valve apparatus. Marker based strain measurements can be performed, but chordae heterogeneity limits utility. Foil-based strain sensors have also been described, but tend to be bulky, limiting the number of chordae that can be measured. Hypothesis: We hypothesize that the use of Fiber Bragg Grating (FBG) sensors—optical strain gauges made of 125 μ m diameter silica fibers— can overcome the critical limitations of previous methods of measuring chordae tendineae forces. Methods: Using FBG sensors, we created a force-sensing neochord that would mimic the natural shape and movement of native chordae tendineae. FBG sensors reflect a specific wavelength of light depending on the spatial period of gratings - when a force is applied, the gratings move relative to one another, changing the reflected light (Fig 1A). This wavelength shift is directly proportional to the force applied. The FBG sensors were housed in a protective sheath fashioned from the outer coil of a 0.025" Amplatz Extra-Stiff guidewire, and attached to the chordae using GoreTex suture (Fig 1B). The function of the force-sensing neochordae were validated in a 3D printed left heart simulator. Results: FBG sensors provided high-fidelity force measurements of mitral valve chordae tendineae at a temporal resolution of 1000 Hz. As ventricular pressures increased, such as in the hypertensive condition, forces on the chordae also increased (Fig 1C). The resolution of FBG sensors allow for increased accuracy of not only chordae tension, but also the rate of change of force (dF/dt) - a parameter critical to determining likelihood of leaflet rupture. Conclusions: FBG sensors are a viable, durable, and high-fidelity technology that can be effectively used to measure mitral valve chordae forces and overcome limitations of other such technologies.