More Like this

1. Act the Part: Learning Interaction Strategies for Articulated Object Part Discovery

   Gadre, Samir Yitzhak; Ehsani, Kiana; Song, Shuran (October 2021, Proceedings)

   People often use physical intuition when manipulating articulated objects, irrespective of object semantics. Motivated by this observation, we identify an important embodied task where an agent must play with objects to recover their parts. To this end, we introduce Act the Part (AtP) to learn how to interact with articulated objects to discover and segment their pieces. By coupling action selection and motion segmentation, AtP is able to isolate structures to make perceptual part recovery possible without semantic labels. Our experiments show AtP learns efficient strategies for part discovery, can generalize to unseen categories, and is capable of conditional reasoning for the task. Although trained in simulation, we show convincing transfer to real world data with no fine-tuning. A summery video, interactive demo, and code will be available at atp.cs.columbia.edu. 

   more » « less
2. Composable Part-Based Manipulation

   Liu, Weiyu; Mao, Jiayuan; Hsu, Joy; Hermans, Tucker; Garg, Animesh; Wu, Jiajun (November 2023, Conference on Robot Learning (CoRL))
3. CUTTING A PART FROM MANY MEASURES

   https://doi.org/10.1017/fms.2019.33  

   BLAGOJEVIĆ, PAVLE V.; PALIĆ, NEVENA; SOBERÓN, PABLO; ZIEGLER, GÜNTER M. (January 2019, Forum of Mathematics, Sigma)

   Holmsen, Kynčl and Valculescu recently conjectured that if a finite set $$X$$ with $$\ell n$$ points in $$\mathbb{R}^{d}$$ that is colored by $$m$$ different colors can be partitioned into $$n$$ subsets of $$\ell$$ points each, such that each subset contains points of at least $$d$$ different colors, then there exists such a partition of $$X$$ with the additional property that the convex hulls of the $$n$$ subsets are pairwise disjoint. We prove a continuous analogue of this conjecture, generalized so that each subset contains points of at least $$c$$ different colors, where we also allow $$c$$ to be greater than $$d$$ . Furthermore, we give lower bounds on the fraction of the points each of the subsets contains from $$c$$ different colors. For example, when $$n\geqslant 2$$ , $$d\geqslant 2$$ , $$c\geqslant d$$ with $$m\geqslant n(c-d)+d$$ are integers, and $$\unicode[STIX]{x1D707}_{1},\ldots ,\unicode[STIX]{x1D707}_{m}$$ are $$m$$ positive finite absolutely continuous measures on $$\mathbb{R}^{d}$$ , we prove that there exists a partition of $$\mathbb{R}^{d}$$ into $$n$$ convex pieces which equiparts the measures $$\unicode[STIX]{x1D707}_{1},\ldots ,\unicode[STIX]{x1D707}_{d-1}$$ , and in addition every piece of the partition has positive measure with respect to at least $$c$$ of the measures $$\unicode[STIX]{x1D707}_{1},\ldots ,\unicode[STIX]{x1D707}_{m}$$ . 

   more » « less
4. Plunging breakers. Part 2. Droplet generation

   https://doi.org/10.1017/jfm.2023.378  

   Erinin, M.A.; Liu, C.; Wang, S.D.; Liu, X.; Duncan, J.H. (July 2023, Journal of Fluid Mechanics)

   An experimental study of the dynamics and droplet production in three mechanically generated plunging breaking waves is presented in this two-part paper. In the present paper (Part 2), in-line cinematic holography is used to measure the positions, diameters ($$d\geq 100\ \mathrm {\mu }{\rm m}$$), times and velocities of droplets generated by the three plunging breaking waves studied in Part 1 (Erininet al.,J. Fluid Mech., vol. 967, 2023, A35) as the droplets move up across a horizontal measurement plane located just above the wave crests. It is found that there are four major mechanisms for droplet production: closure of the indentation between the top surface of the plunging jet and the splash that it creates, the bursting of large bubbles that were entrapped under the plunging jet at impact, splashing and bubble bursting in the turbulent zone on the front face of the wave and the bursting of small bubbles that reach the water surface at the crest of the non-breaking wave following the breaker. The droplet diameter distributions for the entire droplet set for each breaker are fitted with power-law functions in separate small- and large-diameter regions. The droplet diameter where these power-law functions cross increases monotonically from 820 to 1480$$\mathrm {\mu }{\rm m}$$from the weak to the strong breaker, respectively. The droplet diameter and velocity characteristics and the number of the droplets generated by the four mechanisms are found to vary significantly and the processes that create these differences are discussed. 

   more » « less
5. Mutual Learning: Part II --Reinforcement Learning

   https://doi.org/10.23919/ACC45564.2020.9147838  

   Narendra, Kumpati S.; Mukhopadhyay, Snehasis (July 2020, 2020 American Control Conference (ACC))