Learning Deep Policies for Robot Bin Picking by Simulating Robust Grasping Sequences

By Jeffrey Mahler, Ken Goldberg. 2017

202 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.

Citations in the corpus

Deep Imitative Models for Flexible Inference, Planning, and Control (2018, methodology): “A common straightforward approach to IL is behavior-cloning: learning to predict the actions taken by a demonstrator (Pomer-leau, 1989; Mahler & Goldberg, 2017; Codevilla et al., 2018).”
Causal Confusion in Imitation Learning (2019, background): “Imitation learning through behavioral cloning dates back to Widrow and Smith, 1964 [58], and has remained popular through today [41, 44, 36, 37, 4, 13, 33, 56, 3].”
DeceptionNet: Network-Driven Domain Randomization (2019, unknown): “For other approaches, such as robotic control learning, solutions must be found by exploration in tight simulation-based feedback loops that require synthetic rendering [28, 39, 31].”
Learning robust, real-time, reactive robotic grasping (2019, methodology): “…and process grasp candidates using a sliding window at discrete intervals of offset and rotation (Lenz et al., 2015; Pinto and Gupta, 2016), which is computationally expensive and results in grasp planning times of the order of a second (Mahler et al., 2017) to tens of seconds (Lenz et al., 2015).”
“Good Robot!”: Efficient Reinforcement Learning for Multi-Step Visual Tasks with Sim to Real Transfer (2019, background): “tasks with multiple actions: [36] includes placing one block on another, [37] places one towel on a bar, and [38] clears a bin, but the first two are not long-horizon tasks and the possibility of progress reversal (Fig.”
Learning ambidextrous robot grasping policies (2019, background): “0 base networks on simulated heaps with imitation learning from an algorithmic supervisor (26).”
GPU-Accelerated Robotic Simulation for Distributed Reinforcement Learning (2018, background): “Given the recent successes of sim2real transfer learning, from grasping in clutter [28], quadruped locomotion [29], to dexterous manipulation [30], all of which used Bullet to generate simulation data to train policies that worked in the real world, we believe our simulator can provide valuable spee”
6-DOF Grasping for Target-driven Object Manipulation in Clutter (2019, background): “In that regard, it is similar to other data-driven methods for clutter ([11], [13], [14], [29],”