CASSL: Curriculum Accelerated Self-Supervised Learning

By Adithyavairavan Murali, Lerrel Pinto, Dhiraj Gandhi, Abhinav Kumar Gupta. 2017

37 citations recorded by the source corpus.

Recent self-supervised learning approaches focus on using a few thousand data points to learn policies for high-level, low-dimensional action spaces. However, scaling this framework for higher-dimensional control requires either scaling up the data collection efforts or using a clever sampling strategy for training. We present a novel approach - Curriculum Accelerated Self-Supervised Learning (CASSL) - to train policies that map visual information to high-level, higher-dimensional action spaces. CASSL orders the sampling of training data based on control dimensions: the learning and sampling are focused on few control parameters before other parameters. The right curriculum for learning is suggested by variance-based global sensitivity analysis of the control space. We apply our CASSL framework to learning how to grasp using an adaptive, underactuated multi-fingered gripper, a challenging system to control. Our experimental results indicate that CASSL provides significant improvement and generalization compared to baseline methods such as staged curriculum learning (8% increase) and complete end-to-end learning with random exploration (14% improvement) tested on a set of novel objects.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Control and Systems Engineering

How researchers see this paper

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

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Control and Systems Engineering

Citations in the corpus

Multiple Interactions Made Easy (MIME): Large Scale Demonstrations Data for Imitation (2018, background): “This has been followed by attempts to scale even further using multiple robots [3], multiple tasks [19], adversarial learning [20] and curriculum learning [21].”
RODE: Learning Roles to Decompose Multi-Agent Tasks (2020, methodology): “d transfer policy across different tasks (Chen et al., 2019). By contrast, we use action representations to factor multi-agent tasks. Besides learning action representations, Czarnecki et al. (2018); Murali et al. (2018); Farquhar et al. (2020) propose to use curriculum learning over action spaces o”
Robot Learning in Homes: Improving Generalization and Reducing Dataset Bias (2018, background): “Therefore, there has concurrently been a push in the robotics community to collect real-world physical interaction data [4, 5, 6, 7, 8, 9, 10] in multiple robotics labs.”
6-DOF Grasping for Target-driven Object Manipulation in Clutter (2019, methodology): “Some recent methods tackle the more difficult problem of generating grasps in SE(3) from 2D (image) [17], 2.”
Learning to Manipulate Deformable Objects without Demonstrations (2019, background): “This has provided benefits in several other learning problems from generating images [64] to predicting large dimensional robotic actions [40, 62].”
Learning Dexterous Grasping with Object-Centric Visual Affordances (2020, background): “trained to detect successful grasps by parallel jaw grippers can accelerate the robot’s learning [22], [40], [35], [23], [26], [31], [36].”
Growing Action Spaces (2019, methodology): “In a self-supervised control setting, Murali et al. (2018) use sensitivity analysis to automatically define a curriculum over action dimensions and prioritise their search space.”
Evolutionary Population Curriculum for Scaling Multi-Agent Reinforcement Learning (2020, unknown): “…et al., 2009) and much beyond: Karras et al. (2017) propose to progressively increase the network capacity for synthesizing high quality images; Murali et al. (2018) apply a curriculum over the control space for robotic manipulation tasks; several works (Wu & Tian, 2016; Pinto et al., 2017;…”