Shirin Joshi

About

Shirin Joshi is a roboticist whose research lies at the intersection of computer vision, deep learning, and robotic manipulation. Her most impactful work centers on enabling robots to autonomously grasp and manipulate unknown objects in unstructured environments. She is best known for developing the Generative Residual Convolutional Neural Network (GR-ConvNet) and its improved version, GR-ConvNet v2—real-time, multi-grasp detection networks that generate robust antipodal grasps directly from n-channel images. These contributions, cited over 70 times collectively, have significantly advanced the field of robotic grasping. Beyond grasp detection, Joshi has tackled the challenge of multi-step manipulation tasks, proposing novel reinforcement learning frameworks that combine task-progress-based reward shaping with visual planning. Her work on learning manipulation policies from visual observation and Q-value predictions addresses the complexities of long-horizon tasks involving progress reversal. With a growing citation record and a focus on bridging perception and action, Shirin Joshi is establishing herself as a rising voice in autonomous robotic manipulation.

Research Focus

Key Achievements

Top Papers

1. 1
   GR-ConvNet v2: A Real-Time Multi-Grasp Detection Network for Robotic Grasping

   51 citations · 2022
2. 2
   Antipodal Robotic Grasping using Generative Residual Convolutional Neural Network

   21 citations · 2020
3. 3
   Learning Robotic Manipulation Tasks via Task Progress Based Gaussian Reward and Loss Adjusted Exploration

   13 citations · 2021
4. 4
   Robotic Grasping using Deep Reinforcement Learning

   8 citations · 2020
5. 5
   Learning Robotic Manipulation Tasks through Visual Planning.

   3 citations · 2021
6. 6
   Learning Multi-step Robotic Manipulation Policies from Visual Observation of Scene and Q-value Predictions of Previous Action

   2 citations · 2022

Key Collaborators

Contact & Links