Papers by Henrik Iskov Christensen
Surgical Endoscopy and Other Interventional Techniques, Mar 2, 2023
Background No platform for objective, synchronous and on-line evaluation of both intraoperative e... more Background No platform for objective, synchronous and on-line evaluation of both intraoperative error and surgeon physiology yet exists. Electrokardiogram (EKG) metrics have been associated with cognitive and affective features that are known to impact surgical performance but have not yet been analyzed in conjunction with real-time error signals using objective, real-time methods. Methods EKGs and operating console point-of-views (POVs) for fifteen general surgery residents and five non-medically trained participants were captured during three simulated robotic-assisted surgery (RAS) procedures. Time and frequencydomain EKG statistics were extracted from recorded EKGs. Intraoperative errors were detected from operating console POV videos. EKG statistics were synchronized with intraoperative error signals. Relative to personalized baselines, IBI, SDNN and RMSSD decreased 0.15% (S.E. 3.603e-04; P = 3.25e-05), 3.08% (S.E. 1.603e-03; P < 2e-16) and 1.19% (S.E. 2.631e-03; P = 5.66e-06), respectively, during error. Relative LF RMS power decreased 1.44% (S.E. 2.337e-03; P = 8.38e-10), and relative HF RMS power increased 5.51% (S.E. 1.945e-03; P < 2e-16). Conclusions Use of a novel, on-line biometric and operating room data capture and analysis platform enabled detection of distinct operator physiological changes during intraoperative errors. Monitoring operator EKG metrics during surgery may help improve patient outcomes through real-time assessments of intraoperative surgical proficiency and perceived difficulty as well as inform personalized surgical skills development.
Journal of Aerospace Computing Information and Communication, Dec 1, 2007
Autonomous navigation in urban environments inevitably leads to having to switch between various,... more Autonomous navigation in urban environments inevitably leads to having to switch between various, sometimes conflicting control tasks. Sting Racing, a collaboration between Georgia Tech and SAIC, has developed a modular control architecture for this purpose and this paper describes the operation and definition of this architecture through so-called nested hybrid automata. We show how to map the requirements associated with the DARPA Urban Grand Challenge onto these nested automata and illustrate their operation through a number of experimental results.
Pre-training is transformative in supervised learning: a large network trained with large and exi... more Pre-training is transformative in supervised learning: a large network trained with large and existing datasets can be used as an initialization when learning a new task. Such initialization speeds up convergence and leads to higher performance. In this paper, we seek to understand what the formalization for pre-training from only existing and observational data in Reinforcement Learning (RL) is and whether it is possible. We formulate the setting as Batch Meta Reinforcement Learning. We identify MDP mis-identification to be a central challenge and motivate it with theoretical analysis. Combining ideas from Batch RL and Meta RL, we propose tiMe, which learns distillation of multiple value functions and MDP embeddings from only existing data. In challenging control tasks and without additional exploration on unseen MDPs, tiMe is competitive with state-of-the-art model-free RL method trained with hundreds of thousands of interactions. This work demonstrates that Meta RL from observational data is possible and we hope it will gather additional interest from the community to tackle this problem.
Advances in intelligent systems and computing, 2013
Auction based algorithms offer effective methods for de-centralized task assignment in multi-agen... more Auction based algorithms offer effective methods for de-centralized task assignment in multi-agent teams. Typically there is an implicit assumption that agents can be trusted to effectively perform assigned tasks. However, reliable performance of team members may not always be a valid assumption. An approach to learning team member performance is presented, which enables more efficient task assignment. A policy gradient reinforcement learning algorithm is used to learn a cost factor that can be applied individually to auction bids. Experimental results demonstrate that agents that model team member performance using this approach can more efficiently distribute tasks in multi-agent auctions.
Ubiquitous Computing, Sep 16, 2007
Robots have entered our domestic lives, but yet, little is known about their impact on the home. ... more Robots have entered our domestic lives, but yet, little is known about their impact on the home. This paper takes steps towards addressing this omission, by reporting results from an empirical study of iRobot's Roomba™, a vacuuming robot. Our findings suggest that, by developing intimacy to the robot, our participants were able to derive increased pleasure from cleaning, and expended effort to fit Roomba into their homes, and shared it with others. These findings lead us to propose four design implications that we argue could increase people's enthusiasm for smart home technologies.
Springer eBooks, 2022
In recent years, various state of the art autonomous vehicle systems and architectures have been ... more In recent years, various state of the art autonomous vehicle systems and architectures have been introduced. These methods include planners that depend on high-definition (HD) maps and models that learn an autonomous agent's controls in an end-to-end fashion. While end-to-end models are geared towards solving the scalability constraints from HD maps, they do not generalize for different vehicles and sensor configurations. To address these shortcomings, we introduce an approach that leverages lightweight map representations, explicitly enforcing geometric constraints, and learns feasible trajectories using a conditional generative model. Additional contributions include a new dataset that is used to verify our proposed models quantitatively. The results indicate low relative errors that can potentially translate to traversable trajectories. The dataset created as part of this work has been made available online. 1
The present paper examines minimum jerk models for human kinematics as a tool to predict user inp... more The present paper examines minimum jerk models for human kinematics as a tool to predict user input in teleoperation with significant dynamics. Predictions of user input can be a powerful tool to bridge time-delays and to trigger autonomous sub-sequences. In this paper an example implementation is presented, along with the results of a pilot experiment in which a virtual reality simulation of a teleoperated ball-catching scenario is used to test the predictive power of the model. The results show that delays up to 100 ms can potentially be bridged with this approach.
arXiv (Cornell University), Oct 14, 2020
Calibration of sensors is fundamental to robust performance for intelligent vehicles. In natural ... more Calibration of sensors is fundamental to robust performance for intelligent vehicles. In natural environments, disturbances can easily challenge calibration. One possibility is to use natural objects of known shape to recalibrate sensors. An approach based on recognition of traffic signs, such as stop signs, and use of them for recalibration of cameras is presented. The approach is based on detection, geometry estimation, calibration, and recursive updating. Results from natural environments are presented that clearly show convergence and improved performance.
arXiv (Cornell University), Oct 19, 2016
We propose a Convolutional Neural Network (CNN) based algorithm -StuffNet -for object detection. ... more We propose a Convolutional Neural Network (CNN) based algorithm -StuffNet -for object detection. In addition to the standard convolutional features trained for region proposal and object detection , StuffNet uses convolutional features trained for segmentation of objects and 'stuff' (amorphous categories such as ground and water). Through experiments on Pascal VOC 2010, we show the importance of features learnt from stuff segmentation for improving object detection performance. StuffNet improves performance from 18.8% mAP to 23.9% mAP for small objects. We also devise a method to train StuffNet on datasets that do not have stuff segmentation labels. Through experiments on Pascal VOC 2007 and 2012, we demonstrate the effectiveness of this method and show that StuffNet also significantly improves object detection performance on such datasets.
With the recent development of autonomous vehicle technology, there have been active efforts on t... more With the recent development of autonomous vehicle technology, there have been active efforts on the deployment of this technology at different scales that include urban and highway driving. While many of the prototypes showcased have been shown to operate under specific cases, little effort has been made to better understand their shortcomings and generalizability to new areas. Distance, uptime and number of manual disengagements performed during autonomous driving provide a high-level idea on the performance of an autonomous system but without proper data normalization, testing location information, and the number of vehicles involved in testing, the disengagement reports alone do not fully encompass system performance and robustness. Thus, in this study a complete set of metrics are applied for benchmarking autonomous vehicle systems in a variety of scenarios that can be extended for comparison with human drivers and other autonomous vehicle systems. These metrics have been used to benchmark UC San Diegos autonomous vehicle platforms during early deployments for micro-transit and autonomous mail delivery applications.
It has long been recognized that novelty effects exist in the interaction with technologies. De s... more It has long been recognized that novelty effects exist in the interaction with technologies. De spite this recognition, we still know little about the novelty effects associated with domestic robotic appliances and more impor tantly, what occurs after the novelty wears off. To address this gap, we undertook a longitudinal field study with 30 households to which we gave Roomba vacuuming robots and then observed use over six months. During this study, which spans over 149 home visits, we encountered methodological challenges in understanding households' usage patterns. In this paper we report on our longitudinal research, focusing particularly on the methods that we used 1) to understand hum an-robot interaction over time despite the constraints of privacy and tem porality in the hom e, and 2) to uncover inform ation when routines became less conscious to the participants themselves.
arXiv (Cornell University), Sep 19, 2016
arXiv (Cornell University), Apr 11, 2016
Our lives have been immensely improved by decades of automation research - we are more comfortabl... more Our lives have been immensely improved by decades of automation research - we are more comfortable, more productive and safer than ever before. Just imagine a world where familiar automation technologies have failed. In that world, thermostats don't work ----you have to monitor your home heating system manually. Cruise control for your car doesn't exist. Every elevator has to have a human operator to hit the right floor, most manufactured products are assembled by hand, and you have to wash your own dishes. Who would willingly adopt that world -the world of the last century ----today? Physical systems -elevators, cars, home appliances, manufacturing equipment ----were more troublesome, more time-consuming, less safe, and far less convenient. Now, suppose we put ourselves in the place of someone 20 years in the future, a future of autonomous systems. A future where transportation is largely autonomous, more efficient, and far safer; a future where dangerous occupations like mining or disaster response are performed by autonomous systems supervised remotely by humans; a future where manufacturing and healthcare are twice as productive per person--hour by having smart monitoring and readily re--tasked autonomous physical agents; a future where the elderly and infirm have 24 hour in--home autonomous support for the basic activities, both physical and social, of daily life. In a future world where these capabilities are commonplace, why would someone come back to today's world where someone has to put their life at risk to do a menial job, we lose time to mindless activities that have no intrinsic value, or be consumed with worry that a loved one is at risk in their own home?
We address the problem of dispersing a large number of autonomous mobile robots toward building w... more We address the problem of dispersing a large number of autonomous mobile robots toward building wireless ad hoc sensor networks performing environmental monitoring and control. For the purpose, we propose the adaptive triangular mesh generation algorithm that enables robots to generate triangular meshes of various sizes adapting to changing environmental conditions. A locally interacting, geometric technique allows robots to generate each triangular mesh with their two neighbor robots. Specifically, we have assumed that robots are not allowed to have the identifier, any pre-determined leaders or common coordinate systems, and any explicit communication. Under such minimal conditions, the positions of the robots were shown to converge to the desired distribution, which was mathematically proven and also verified through extensive simulations. Our preliminary results indicate that the proposed algorithm can be applied to the problem regarding the coverage of an area of interest by a swarm of mobile sensors.
A key aspect of service robotics for everyday use is the motion of systems in close proximity to ... more A key aspect of service robotics for everyday use is the motion of systems in close proximity to humans. It is here essential that the robot exhibits a behaviour that signals safe motion and awareness of the other actors in its environment. To facilitate this there is a need to endow the system with facilities for detection and tracking of objects in the vicinity of the platform, and to design a control law that enables motion generation which is considered socially acceptable. We present a system for in-door navigation in which the rules of proxemics are used to define interaction strategies for the platform.
Least squares minimization of the differential epipolar constraint is a fast and efficient techni... more Least squares minimization of the differential epipolar constraint is a fast and efficient technique to estimate structure and motion for pair of views. Previous work in this area showed how unbiased and consistent estimates could be obtained minimizing the squared errors. However, it implicitly assumes that the errors along the x and y directions are identical and uncorrelated. This is rarely the case for real data, due to the aperture problem. Instead, one should minimize the covariance weighted squared error. Moreover, when dense sequences are acquired, further robustness can be achieved by integrating the reconstruction of structure over time. This paper has two main contributions: (i) we show that the minimization of the weighted squared errors (i.e. Maximum-Likelihood estimate) outperforms the more traditional approach of un-weighted least squares, (ii) we show how structure estimation can be integrated over time in a multi-view approach that drastically improves estimates.
arXiv (Cornell University), Mar 27, 2019
IEEE robotics and automation letters, Oct 1, 2019
Event data recording is crucial in robotics research, providing prolonged insights into a robot's... more Event data recording is crucial in robotics research, providing prolonged insights into a robot's situational understanding, progression of behavioral state, and resulting outcomes. Such recordings are invaluable when debugging complex robotic applications or profiling experiments ex post facto. As robotic developments mature into production, both the roles and requirements of event logging will broaden, to include serving as evidence for auditors and regulators investigating accidents or fraud. Given the growing number of high profile public incidents involving selfdriving automotives resulting in fatality and regulatory policy making, it is paramount that the integrity, authenticity and nonrepudiation of such event logs are maintained to ensure accountability. Being mobile cyber-physical systems, robots present new threats and vulnerabilities beyond traditional IT: unsupervised physical system access or postmortem collusion between robot and OEM could result in the truncation or alteration of prior records. In this work, we address immutablization of log records via integrity proofs and distributed ledgers with special considerations for mobile and public service robot deployments.
Uploads
Papers by Henrik Iskov Christensen