Ioannis Poulakakis

Ioannis Poulakakis, Luca Scardovi, Naomi Ehrich Leonard

This paper considers a network of stochastic evidence accumulators, each represented by a drift-diffusion model accruing evidence towards a decision in continuous time by observing a noisy signal and by exchanging information with other units according to a fixed communication graph. We bring into focus the relationship between the location of each unit in the communication graph and its certainty as measured by the inverse of the variance of its state. We show that node classification according to degree distributions or geodesic distances cannot faithfully capture node ranking in terms of certainty. Instead, all possible paths connecting each unit with the rest in the network must be incorporated. We make this precise by proving that node classification according to information centrality provides a rank ordering with respect to node certainty, thereby affording a direct interpretation of the certainty level of each unit in terms of the structural properties of the underlying communication graph.

Sushant Veer, Rakesh, Ioannis Poulakakis · princeton

In this paper we investigate the relation between robustness of periodic orbits exhibited by systems with impulse effects and robustness of their corresponding Poincaré maps. In particular, we prove that input-to-state stability (ISS) of a periodic orbit under external excitation in both continuous and discrete time is equivalent to ISS of the corresponding 0-input fixed point of the associated \emph{forced} Poincaré map. This result extends the classical Poincaré analysis for asymptotic stability of periodic solutions to establish orbital input-to-state stability of such solutions under external excitation. In our proof, we define the forced Poincaré map, and use it to construct ISS estimates for the periodic orbit in terms of ISS estimates of this map under mild assumptions on the input signals. As a consequence of the availability of these estimates, the equivalence between exponential stability (ES) of the fixed point of the 0-input (unforced) Poincaré map and ES of the corresponding orbit is recovered. The results can be applied naturally to study the robustness of periodic orbits of continuous-time systems as well. Although our motivation for extending classical Poincaré analysis to address ISS stems from the need to design robust controllers for limit-cycle walking and running robots, the results are applicable to a much broader class of systems that exhibit periodic solutions.

Abhijeet M. Kulkarni, Ioannis Poulakakis, Guoquan Huang

Proprioceptive-only state estimation is attractive for legged robots since it is computationally cheaper and is unaffected by perceptually degraded conditions. The history of joint-level measurements contains rich information that can be used to infer the dynamics of the system and subsequently produce navigational measurements. Recent approaches produce these estimates with learned measurement models and fuse with IMU data, under a Gaussian noise assumption. However, this assumption can easily break down with limited training data and render the estimates inconsistent and potentially divergent. In this work, we propose a proprioceptive-only state estimation framework for legged robots that characterizes the measurement noise using set-coverage statements that do not assume any distribution. We develop a practical and computationally inexpensive method to use these set-coverage measurements with a Gaussian filter in a systematic way. We validate the approach in both simulation and two real-world quadrupedal datasets. Comparison with the Gaussian baselines shows that our proposed method remains consistent and is not prone to drift under real noise scenarios.

Prem Chand, Sushant Veer, Ioannis Poulakakis

In this paper, we consider the problem of adapting a dynamically walking bipedal robot to follow a leading co-worker while engaging in tasks that require physical interaction. Our approach relies on switching among a family of Dynamic Movement Primitives (DMPs) as governed by a supervisor. We train the supervisor to orchestrate the switching among the DMPs in order to adapt to the leader's intentions, which are only implicitly available in the form of interaction forces. The primary contribution of our approach is its ability to furnish certificates of generalization to novel leader intentions for the trained supervisor. This is achieved by leveraging the Probably Approximately Correct (PAC)-Bayes bounds from generalization theory. We demonstrate the efficacy of our approach by training a neural-network supervisor to adapt the gait of a dynamically walking biped to a leading collaborator whose intended trajectory is not known explicitly.

Sushant Veer, Ioannis Poulakakis

Complex motions for robots are frequently generated by switching among a collection of individual movement primitives. We use this approach to formulate robot motion plans as sequences of primitives to be executed one after the other. When dealing with dynamical movement primitives, besides accomplishing the high-level objective, planners must also reason about the effect of the plan's execution on the safety of the platform. This task becomes more daunting in the presence of disturbances, such as external forces. To alleviate this issue, we present a framework that builds on rigorous control-theoretic tools to generate safely-executable motion plans for externally excited robotic systems. Our framework is illustrated on a 3D limit-cycle gait bipedal robot that adapts its walking pattern to persistent external forcing.

Sushant Veer, Ioannis Poulakakis

In this paper, we investigate the robustness to external disturbances of switched discrete and continuous systems with multiple equilibria. It is shown that if each subsystem of the switched system is Input-to-State Stable (ISS), then under switching signals that satisfy an average dwell-time bound, the solutions are ultimately bounded within a compact set. Furthermore, the size of this set varies monotonically with the supremum norm of the disturbance signal. It is observed that when the subsystems share a common equilibrium, ISS is recovered for solutions of the corresponding switched system; hence, the results in this paper are a natural generalization of classical results in switched systems that exhibit a common equilibrium. Additionally, we provide a method to analytically compute the average dwell time if each subsystem possesses a quadratic ISS-Lyapunov function. Our motivation for studying this class of switched systems arises from certain motion planning problems in robotics, where primitive motions, each corresponding to an equilibrium point of a dynamical system, must be composed to realize a task. However, the results are relevant to a much broader class of applications, in which composition of different modes of behavior is required.

Chetan D. Pahlajani, Ioannis Poulakakis, Herbert G. Tanner

This paper addresses a detection problem where several spatially distributed sensors independently observe a time-inhomogeneous stochastic process. The task is to decide between two hypotheses regarding the statistics of the observed process at the end of a fixed time interval. In the proposed method, each of the sensors transmits once to a fusion center a locally processed summary of its information in the form of a likelihood ratio. The fusion center then combines these messages to arrive at an optimal decision in the Neyman-Pearson framework. The approach is motivated by applications arising in the detection of mobile radioactive sources, and offers a pathway toward the development of novel fixed- interval detection algorithms that combine decentralized processing with optimal centralized decision making.