Automated Construction of Robotic Manipulation Programs

By Rosen Diankov. 2010

622 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Reinforcement Learning

Citations in the corpus

Learning Trajectory Preferences for Manipulators via Iterative Improvement (2013, neutral): “edback mechanisms that enable the user to easily provide improved trajectories. (a)Re-ranking: We rank trajectories in order of their current predicted scores and visualize the ranking using OpenRave [7]. User observers trajectories sequentially and clicks on the ﬁrst trajectory which is better than”
Learning human activities and object affordances from RGB-D videos (2012, methodology): “ Cornell’s Kodiak, a PR2 robot, in our experiments. Kodiak is mounted with a Microsoft Kinect, which is used as the main input sensor to obtain the RGB-D video stream. We used the OpenRAVE libraries (Diankov, 2010) for programming the robot to perform the pre-programmed assistive tasks. Assisting Hu”
Learning Synergies Between Pushing and Grasping with Self-Supervised Deep Reinforcement Learning (2018, methodology): “Other methods [28], [6] explore the model-free planning of pushing motions to move objects to target positions that are more favorable for pre-designed grasping algorithms – the behaviors of which are typically handcrafted, fixed, and well-known in advance.”
Supersizing self-supervision: Learning to grasp from 50K tries and 700 robot hours (2015, background): “[13], [19], [20] offer reasons as to why a simulated environment and an analytic metric would not parallel the real world which is highly unstructured.”
Data-Driven Grasp Synthesis—A Survey (2013, background): “Diankov [20] claims that in practice grasps synthesized using this metric tend to be relatively fragile .”
TossingBot: Learning to Throw Arbitrary Objects With Residual Physics (2019, methodology): “the gripper ﬁngertips meets x, at which point the gripper closes, and lifts upwards 10cm. This primitive is open-loop, with robot arm motion planning executed using a stable, collision-free IK solver [8]. Grasping network. The grasping network is a 7-layer fully convolutional residual network [4,14,”
Anticipating Human Activities Using Object Affordances for Reactive Robotic Response (2013, methodology): “We used the OpenRAVE libraries [2] for programing the robot to perform the preprogrammed tasks described in the aforementioned scenarios by incorporating the anticipations generated with our ATCRFs.”
Whole-body motion planning for manipulation of articulated objects (2013, methodology): “To analytically solve IK, we use IKfast [25].”
Form2Fit: Learning Shape Priors for Generalizable Assembly from Disassembly (2019, methodology): “This primitive executes in an open loop fashion using stable, collision-free IK solves [23].”
Integrated Task and Motion Planning (2020, methodology): “) either by projecting random samples into the constraint set using optimization (41) or by analytically solving for the solutions to a reparameterized set that captures its underlying dimensionality (42). Figure 4 demonstrates a 1D robot (R) acting in the presence of a single movable object (A). Th”
Pick and Place Without Geometric Object Models (2017, methodology): “p quality metrics [22]. The later can be simulated by evaluating sufﬁcient conditions to determine whether an object will fall over given the executed placement. Both are easy to evaluate in OpenRAVE [23], the simulator used in this work. D. The Action-Value Function We approximate the action-value ”
High precision grasp pose detection in dense clutter (2016, methodology): “We use OpenRAVE [2] for collision checking, and we model the point cloud obstacle as 2cm cubes centered at points in a voxelized cloud.”
Combined task and motion planning through an extensible planner-independent interface layer (2014, methodology): “We implemented the proposed approach using the OpenRAVE simulator [4].”
Online Replanning in Belief Space for Partially Observable Task and Motion Problems (2019, methodology): “ow task, the robot must remove the sugar to place the block and close the drawer. VIII. EXPERIMENTS within 4 simulated domains. We used PyBullet [36] for ray-casting and collision checking and IKFast [37] for inverse kinematics. See https://github.com/caelan/ SS-Replan for our open-source Python imp”
Learning Visual Affordances for Robotic Manipulation (2019, methodology): “In both primitives, robot arm motion planning is automatically executed with stable, collision-free IK solves [22].”
Robotic pick-and-place of novel objects in clutter with multi-affordance grasping and cross-domain image matching (2017, positive): “with stable inverse kinematic-based controllers (Diankov, 2010).”
Leveraging big data for grasp planning (2015, methodology): “-metric has been deemed unreliable [6, 17, 37].”