Utkarsh Mishra
@utkarshm0410
Followers
329
Following
810
Media
18
Statuses
65
Robotics PhD Student @GeorgiaTech || TRI Summer’24 || Robot Learning || IITR'21 || 🎸🎢🤖|| He/Him || views are mine.
Atlanta, GA
Joined July 2015
How can robots compositionally generalize over multi-object multi-robot tasks for long-horizon planning?. At #CoRL2024, we introduce Generative Factor Chaining (GFC), a diffusion-based approach that composes spatial-temporal factors into long-horizon skill plans. (1/7)
2
34
148
RT @saxenavaibhav11: Large robot datasets are crucial for training 🤖foundation models. Yet, we lack systematic understanding of what data m….
0
43
0
RT @lucy_x_shi: If you are interested in solving complex long-horizon tasks, please join us at the 3rd workshop on Learning Effective Abstr….
0
2
0
RT @tomssilver: We're excited to announce the third workshop on LEAP: Learning Effective Abstractions for Planning, to be held at #CoRL2025….
0
15
0
RT @NadunRanawakaA: Tired of slow-moving robots? Want to know how learning-driven robots can move closer to industrial speeds in the real w….
0
8
0
RT @danfei_xu: Despite advances in end-to-end policies, robots powered by these systems operate far below industrial speeds. What will it….
0
3
0
RT @WeiGuo01: How annealing helps overcoming multimodality? In our ICLR 2025 paper and preprint .
0
27
0
RT @simar_kareer: Introducing EgoMimic - just wear a pair of Project Aria @meta_aria smart glasses 👓 to scale up your imitation learning da….
0
54
0
1
0
6
While factor diffusion models are only trained with single-arm manipulation data, GFC readily allows performing collaborative tasks where multiple manipulators operate on the same object. We use this to reorient a pot given each arm factors know how to grasp a pot. (6/7)
1
0
3
GFC allows easy plug-and-play with pre-train factor diffusion models and supports reusing of factors for new tasks. Pouring from one cup to another requires knowinh how to grasp a cup, how to move a cup and feasible spatial relationship between cups for pouring. (5/7)
1
0
3
GFC effectively handles long-horizon dependencies, solving planning tasks with up to 20 skills. Its factor-based design scales across multi-manipulator settings, managing diverse interactions without retraining. We validate this with real-world experiments. (4/7)
1
0
4
During inference, GFC composes factors, adding constraints as needed to create task-level plans from factor distributions, enabling generalization to new tasks. Overlapping distributions ensure feasibility of all spatial and temporal factors. (3/7)
1
0
4
To train GFC, we use diffusion models that learn distribution of each spatial/temporal factor (constraint/skill) independently from state configurations & skills. Temporal factors represent the distribution of (pre-condition, skill parameter, effect) for individual skills. (2/7)
1
0
6
RT @HadasKressGazit: Evaluation in robot learning papers, or, please stop using only success rate.a paper and a 🧵.
arxiv.org
The robot learning community has made great strides in recent years, proposing new architectures and showcasing impressive new capabilities; however, the dominant metric used in the literature,...
0
41
0
RT @Haotianxue_GT: #ICLR2024 **The LDM is Vulnerable because of the Encoder! ** Our paper: Towards More Effective Protection Against Diff….
0
6
0
RT @danfei_xu: Since we are entering the "BC is all you need" phase of Robot Learning😜 --- Robomimic ( allows you t….
0
19
0
RT @YongxinChen1: Delighted to announce the successful defense of my exceptional PhD student @qsh_zh @GTrobotics @mlatgt @gatechengineers S….
0
4
0
RT @nishanthkumar23: There was a lot of good and interesting debate on "is scaling all we need to solve robotics?" at #CoRL23. I spent some….
0
43
0