15 July 2019
|08.30 – 09.15||Registration|
|09.15 – 09.30||Bharadwaj Amrutur||Welcome Address|
|SESSION 1: Socio-technical systems (Session Chair – Rajesh Sundaresan)|
|09.30 – 10.30||Krishna Gummadi||Foundations for fair algorithmic decision making
As algorithms are increasingly used to make important decisions that affect human lives, ranging from predicting risk of criminal recidivism to nudging consumer decisions on online sites, concerns have been raised about the fairness of algorithmic (data-driven and learning-based) decision making. A number of recent works have proposed methods to measure and eliminate discrimination in algorithmic decision outcomes. In this talk, I will argue that the notions of fairness considered in these early works are limited along several dimensions: (i) they focus on distributive fairness (i.e., fairness of the outcomes or ends of decision making) at the expense of procedural fairness (i.e., fairness of the process or means of decision making); (b) they normatively prescribe how fair decisions ought to be made rather than descriptively study how people perceive and reason about fairness of decisions; and (c) they ignore the human-in-the-loop, i.e., how algorithmic predictions influence the decisions of humans who rely on them. I will present a few measures and mechanisms to quantify and mitigate algorithmic unfairness along these previously overlooked dimensions and discuss the challenging tradeoffs that arise when we attempt to account for all the different fairness considerations simultaneously.
|10.30 – 11.00||Sunil Mani||Diffusion of automation technologies and their potential and actual effects on manufacturing employment in India
The anxiety that technology will displace jobs on a large scale in the near future is flooding both academic and public debates, primarily in the developed world. The recent publication of a study by Oxford Martin School has predicted that a large number of occupations will see an increased rate of automation which is likely to have an adverse effect on employment, especially in the manufacturing sector. Employing a comprehensive dataset from the International Federation of Robotics, the study analyses the nature and extent of diffusion industrial robots in Indian manufacturing industry. Instead of an occupation-based approach, the study uses a task-based one, which presents a more accurate picture of the effect of automation on manufacturing employment.
Our analysis shows that the industries that employ robots in the manufacturing operations are the same industries that were historically speaking of using robots. Further, there appears to be a similar stickiness in the tasks that are prone to automation. The paper ends on a cautious note as it is hard to predict cautions that it’s hard to predict the future when it comes to technology. “The line in Silicon Valley is, ‘We always overestimate the amount of change that can occur in a year and we underestimate what can occur in a decade. So in the future increased automation can have a deleterious effect on employment generation.
|11.00 – 11.30||Arnaud Reynaud||Can we nudge farmers into saving water? Evidence from a randomised experiment
We test whether social comparison nudges can promote water-saving behavior among farmers as a complement to traditional CAP measures. We conducted a randomised controlled trial among 200 farmers equipped with irrigation smart meters in South-West France. Treated farmers received weekly information on individual and group water consumption over four months. Our results rule out medium to large effect-sizes of the nudge. Moreover, they suggest that the nudge was effective at reducing the consumption of those who irrigate the most, although it appears to have reduced the proportion of those who do not consume water at all.
|11.30 – 12.00||Break||Ballroom (next to Faculty Hall)|
|SESSION 2: Drones and Autonomous Vehicles (Session Chair – Debasish Ghosh)|
|12.00 – 13.00||C.V. Jawahar||Computer vision for Indian roads: Challenges and Directions
Computer Vision, a sub-area of modern AI, provides capabaility to sense and understand the physical world around, and there by connecting the intelligent computational modules to the real world applications. Indian roads seriously demand technology intervention for improving the safety, navigation and addressing the infrastructural limitations. AI and computer vision hold a lot of potential for improving the situation of the roads in countries like India. In this talk, we discuss how data driven modern AI can help in creating solutions with some specific examples, and list some of the challenges ahead of us.
|13.00 – 14.00||Lunch||Main Guest House|
|14.00 – 14.30||P.B. Sujit||Cooperative UAV-UGV planning for large scale mission
Low cost UAVs are useful in coverage and monitoring applications but need multiple refuels to accomplish large scale mission. In order to provide refueling, one way is to deploy a moving ground vehicle that can cooperate with the UAV to determine locations for refueling, and the other way is to place multiple static refueling stations optimally so that the UAV and UGV can efficiently cover the region. In this talk, I will show cooperative UAV-UGV MILP-based formulations for coverage and monitoring applications along with field experiments involving a UAV and UGV to demonstrate the proof of concept.
|14.30 – 15.00||K. Madhava Krishna||Has modern ML been a game changer for robotics?
That AI-ML (specifically Deep Learning) has energized and propelled the state of the art in Computer Vision, NLP and Speech Processing is well known. Has such post modern ML made it possible to solve previously difficult problems in robotics? This talk will focus on certain specific examples where the pivotal role of ML in pushing the frontiers of robotics is evident. These examples draws upon from research themes that we have been pursuing over the last several years at the Robotics Research Center, IIIT Hyderabad.
|15.00 – 15.30||Break||Ballroom (next to Faculty Hall)|
|15.30 – 17.30||Poster and Demo Session / Industry exhibits||Ballroom (next to Faculty Hall)|
|TUTORIALS (Session Chair – Alexandre Reiffers, Raghu Krishnapuram, Shishir.N.Y)||Department of Electrical Communication Engineering, IISc|
|16.30 – 18.30||Damon Wischik||The capacity of the Gaussian retail channel: How information theory provides the bridge between choice models and resource allocation models
How should Uber set its prices and allocate its drivers? There are two separate types of modelling that are needed. First, we need to model how users make choices and how their choices depend on price; this is the field of discrete choice modelling, a staple of transport modelling pioneered by the Nobel prize winning economist Daniel McFadden. Second, we need to model traffic flows and how capacity should be allocated in a network; this is the field of network utility maximization, pioneered by Frank Kelly in the context of Internet congestion control. It turns out that the two types of modelling can be linked, via a surprising application of Shannon’s information theory.
In this tutorial I will describe the three areas of theory, and show how they come together to answer the Uber question. I will also suggest avenues of research on how machine learning models of human behaviour might be integrated with models of urban or economic systems.
|16.30 – 18.30||C.V. Jawahar||Computer vision for autonomous driving
This tutorial will be organized in two parts. Part I will include an overview of the problem space, associated computer vision tasks, a brief introduction to selected computer vision algorithms, and recent advances that have enabled rapid progress in this area. Part II will address the problem of semantic segmentation of road scenes, algorithms for semantic segmentation, and technical challenges and possible directions.
|16.30 – 18.30||Asokan Thondiyath||Design and control of surgical robots
A broad overview of robotics in health care will be presented with specific focus on surgical applications. Design and control of tele-operated master-slave robotic system for surgical application will be presented and discussed with a case study. The importance of haptic feedback in tele-surgery and the design of haptic feedback systems for surgical robot will also be covered.
16 July 2019
|08.45 – 09.30||Registration|
|SESSION 3: Robotics (Session Chair – Shalabh Bhatnagar)|
|09.30 – 10.30||Jaydev P. Desai||Flexible, 3D-printed robotic systems for surgical interventions
Over the past few decades, robotic systems for surgical interventions have undergone tremendous transformation. The goal of a surgical intervention is to try to do it as minimally invasively as possible, since that significantly reduces post-operative morbidity, reduces recovery time, and also leads to lower healthcare costs. However, minimally invasive surgical interventions for a range of procedures will require a significant change in the healthcare paradigm for both diagnostic and therapeutic interventions. Advances in surgical interventions will benefit from “patient-specific robotic tools” to deliver optimal diagnosis and therapy. Hence, this talk will focus on the development of continuum, flexible, and 3D-printed robotic systems that could be patient-specific. Since, these robotic systems could operate in an imaging environment, we will also address challenges in image-guided interventions. This talk will present examples from neurosurgery and endovascular interventions to highlight the applicability of 3-D printed robotic systems for surgery.
|10.30 – 11.00||Cohan Sujay Carlos and Vaishnavi Parab||Cooking robots
The advent of machines capable of cooking complex dishes end-to-end has opened up new possibilities for the use of automation in the food industry. The design of cooking robots is bound to vary widely according to the quantity of food to be prepared at one go by the system, and the processes involved. However, it is possible to draw some general conclusions concerning the design considerations of such machines. Industrial scale machines need to be able to cook quantities of the order of 20 kg at a go to be usable by businesses running centralised kitchens. Individual machines can be (and today tend to be) highly specialised as there are no space considerations that prevent different machines from being deployed for different types of foods. Cooking robots intended for use in homes need to be able to prepare a larger number of dishes since space is a constraint. They are also at the same time required to be very easy to clean and to operate. In this talk, we discuss a number of practical considerations that influence the design of cooking robots for use in homes. We also examine some physical models and predictive algorithms that can be used in the same.
|11.00 – 11.30||Laxmidhar Behera||Skill learning through programming by demonstration
This talk is concerned with the skill transfer to robotic systems through human demonstrations. In this approach, any complex task is represented in terms of simpler tasks, technically termed as dynamic movement primitives. Human demonstrations are encoded in terms of dynamic systems models such as stochastic regressive dynamical representations. It will be shown that the use of proper Lyapunov functions can lead to the generation of stable robot behaviour through skill transfer. Experimental studies have been carried out to demonstrate efficacy of the proposed approaches in robotic manipulations. These principles have been extended to multiple mobile robot systems to perform cooperative tasks as well.
|11.30 – 12.00||Break||Ballroom (next to Faculty Hall)|
|Industry Session (Session Chair – Raghu Krishnapuram)|
|12.00 – 12.20||Vinod Pathangay (Wipro)||Light-weight visual navigation for autonomous drones
As drones are used in indoor environments which are GPS-denied, the task of navigation has to be accomplished using visual and inertial sensors. In this talk, we explore different high-level approaches for scene understanding from the drone camera along with possible fusion of signals from IMU (inertial measurement unit). With the recently proposed work on visual SLAM (simultaneous localization and mapping) and visual-inertial SLAM, we show that lack of parallax in frames from a monocular camera can lead to intialization issues. We propose a path generation strategy that can overcome the initialization problem and avoid the expensive computation of SLAM. This uses a multi-view coarse depthmap generation for path planning. We show that this approach is relatively light-weight and works for simple environments.
|12.20 – 12.40||Balamuralidhar P. (TCS)||Cognitive robotics
The field of cognitive robotics deals with imparting long term autonomy and human interaction capability to robots through adopting concepts and architectures inspired from human cognitive system. Beyond dealing with complex objects physically in uncertain and challenging environments, a robot also needs to perform linguistic and/or logical tasks. This often requires high-level cognitive capabilities such as logical inference, planning, language and low-level cognitive capability such as physical control, behavioral motion generation and sensory perception. Learning and acquisition of knowledge from multiple levels of perception, also through active interaction with the environment, are important aspects of it. Layers of knowledge need to be represented and integrated towards a world model that drives the underlying cognitive processes. Low level and high level task planning helps to synthesize and orchestrate these processes. Often multiple learning mechanisms including imitation learning and reinforcement learning are adopted using popular machine learning approaches such as deep learning and hierarchical probabilistic learning. This talk will present some of the key trends on this topic and discuss related research from TCS Research & Innovation Labs.
|12.40 – 13.00||Srinivasan Iyengar (Microsoft Research)||Detecting anomalies in solar power generation
The proliferation of solar deployments has significantly increased over the years. Analyzing these deployments can lead to the timely detection of anomalies in power generation, which can maximize the benefits from solar energy. In this work, we present SolarClique, a data-driven approach that can flag anomalies in power generation with high accuracy. Unlike prior approaches, our work neither depends on expensive instrumentation nor does it require external inputs such as weather data. Rather our approach exploits correlations in solar power generation from geographically nearby sites to predict the expected output of a site and flag anomalies. We evaluate our approach on 88 solar installations located in Austin, Texas. We show that our algorithm can even work with data from few geographically nearby sites (>5 sites) to produce results with high accuracy. Thus, our approach can scale to sparsely populated regions, where there are few solar installations. Further, among the 88 installations, our approach reported 76 sites with anomalies in power generation. Moreover, our approach is robust enough to distinguish between reduction in power output due to anomalies and other factors such as cloudy conditions. Additionally, we propose use of drones and thermal cameras to classify the type of anomaly in power generation.
|13.00 – 14.00||Lunch||Main Guest House|
|14.00 – 14.20||Saurabh Chandra (Ati Motors)||Autonomous industrial vehicle
We will give an overview of our product, technologies and showcase a video showing the vehicle in action.
|14.20 – 14.40||Ajay Gurjar (Yaskawa)||Yaskawa Initiatives in AI & Industry 4.0 for Robotics
Yaskawa from top 100 innovative companies has established his footprint in AI with the help of force & vision sensor which help industry to maximize his productivity. Automation and robotics provide the muscle for Industry 4.0, AR/VR, cameras and other sensors provide the senses, and data and connectivity are its central nervous system. But the real brain behind this industrial revolution is AI (Artificial Intelligence).
|14.40 – 15.00||Shanthanu Chakravarthy (Mimyk)||Cyber-Physical systems for training and assistance in medical interventions
Surgeries and interventional procedures are turning into a disease. According to a recent Lancet report, there are more people dying due to surgical complications than HIV, Malaria, and TB combined. At Mimyk, we are working on technologies to improve patient safety and surgical outcomes. This talk will discuss our efforts towards building an immersive platform for training doctors in endoscopy. I will also touch upon our work on in-procedure decision and support system.
|15.00 – 15.30||Break||Ballroom (next to Faculty Hall)|
|TUTORIALS||Department of Electrical Communication Engineering, IISc|
|15.30 – 18.00||Raghu Krishnapuram||Autonomous navigation: Basics of localization and mapping
The first part of this tutorial will cover basics of EKF (Extended Kalman Filter), UKF (Unscented Kalman Filter)and particle filtering, which comprise the foundation of autonomous navigation. The second part of the tutorial will cover the application of these filters to localization and mapping, namely, Markov and Gaussian localization, grid and Monte Carlo localization, occupancy grid mapping, and simultaneous localization and mapping (SLAM) with Rao-Blackwellized particle filters.