Au bâtiment Turing, Ecole Polytechnique: le 26 novembre à partir de 12:00, à l'espace innovation, le 27 novembre matin à l'espace innovation également, jusqu'à 15:00.
Le social event est au Zeyer, à 7:30pm précises, le 26 novembre.
Mardi 26 novembre
12:00am-1:30pm déjeuner - espace innovation
1:30pm-2:15pm Gabriel Betton, Mathis Riera, Loïck Degorre, Lionel Lapierre, Luc Jaulin (ENSTA Bretagne), " A time multiplexing technique to control the heading of an underwater robot using an inertia wheel ", video
Abstract: In this presentation, we give a new control approach to stabilize the heading of an underwater robot with a single inertia wheel. To our knowledge, the controllability of this nonlinear system cannot be proved using existing controllability criteria. Moreover, in our problem, we have several variables to control with a unique scalar input. The principle of our approach is to ask the system to perform a cycle. The shape of the cycle which is tuned using a multiplexing technique that will allow us to stabilize independently all outputs. An illustration on a real underwater robot illustrates the feasibility of the approach.
2:25pm-3:10pm, Franco Cordeiro (Télécom Paris): " Shackling Uncertainty using Mixed Criticality in Monte-Carlo Tree Search ", video
Abstract : In the world of embedded systems, optimizing actions with the uncertain costs of multiple resources in order to achieve an objective is a complex challenge. Existing methods include plan building based on Monte Carlo Tree Search (MCTS), an approach that thrives in multiple online planning scenarios. However, these methods often overlook uncertainty in worst-case cost estimations. A system can fail to operate/function before achieving a critical objective when actual costs exceed optimistic worst-case estimates, even if replanning is considered. Conversely, a system based on pessimistic worst-case estimates would lead to resource over-provisioning even for less critical objectives. To solve similar issues, the Mixed Criticality (MC) approach has been developed in the real-time systems community. We propose an extension of MCTS with MC concepts, which we refer to as (MC)² TS, to efficiently adjust resource allocation to uncertain costs according to the criticality of actions.
3:20pm-3:35pm, Coffee break
3:35pm-4:20pm Christine Tasson (ISAE Supaéro): " Semantics for Reactive Probabilistic Programming ", video
Abstract: Synchronous languages are now a standard industry tool for critical embedded systems. Designers write high-level specifications by composing streams of values using block diagrams. These languages have been extended with Bayesian reasoning to program state-space models which compute a stream of distributions given a stream of observations [1]. This talk aims at describing semantics for probabilistic synchronous languages, based on a joint work with Guillaume Baudart and Louis Mandel [2]. The key idea is to interpret probabilistic expressions as a stream of un-normalized density functions which maps random variable values to a result and positive score. Two equivalent semantics are presented: the co-iterative semantics is executable while the relational semantics is easy to use for proving program equivalence. The semantical framework is applied to prove the correctness of a program transformation required to run an optimized inference algorithm. [1] Reactive Probabilistic Programming, Guillaume Baudart et al, PLDI 2020 [2] Density-Based Semantics for Reactive Probabilistic Programming, Guillaume Baudart, Louis Mandel, Christine Tasson, arxiv:2308.01676
4:30pm-5:15pm, Quentin Brateau*, Fabrice Le Bars, Luc Jaulin (ENSTA Bretagne): " Stable cycles for underwater navigation ", video
Abstract : In GNSS-denied environments, autonomous navigation poses significant challenges, particularly for underwater robots. This work introduces a new paradigm for navigating such environments using stable cycles, allowing the robot to operate without localization. By leveraging prior knowledge of the environment, such as bathymetry, the robot follows a predefined trajectory defined by a timed automaton. The automaton’s durations are adjusted based on a few bathymetric measurements obtained from an echosounder, enabling the stabilization of the robot's cycle on an isobath. The first contribution of this approach is achieving trajectory stabilization along an isobath by controlling the cycle's position. The second contribution involves navigating between stable cycles without GNSS. The robot performs open-loop navigation to approach the capture basin of the next cycle, then switches to a new automaton to stabilize the robot's trajectory around the new stable cycle. This method demonstrates a robust approach to navigation in GNSS-denied environments, ensuring the robot can traverse complex underwater terrains without getting lost.
Mardi 11 juin
9:00am-9:30am: Café-viennoiseries
9:30am-10:15am, Lionel Lapierre (ENSTA Bretagne): " A Subterranean Aquatic Robotic Journey: Robotic Systems for Karst Exploration ", audio
Abstract: Karst landscapes typically consist of an extensive network of underground natural conduits formed through the dissolution of soluble rocks like limestone, dolomite, and gypsum, which can channel large amounts of groundwater. These karst aquifers provide drinking water to millions globally. Making an understanding of karst geomorphology and seasonal variations is essential for ensuring the protection, prediction, and security of this vital resource. Additionally, many transboundary aquifers are contentious, and recent conflicts have highlighted the use of water availability as a weapon in warfare. Assessing the geometry of flow path networks in karst, which governs groundwater dynamics and transport processes, is a challenging scientific goal that relies on field data, often difficult to obtain. While cave divers play a crucial role, they face physiological limitations. Introducing robotic solutions could lead to significant advancements, enabling deeper and more extensive exploration of the karst system. Achieving this requires an interdisciplinary approach, where hydrogeologists, mathematicians, and experts in electronics and control systems collaborate toward a shared objective. This talk will present the latest advancements in robotic exploration of such confined, unstructured, and subaquatic environments, along with the associated challenges.
10:25am-11:10am, (ENSTA Bretagne): " Exploiting Nonlinear Modes for Natural Robotic Locomotion ", video
Abstract: We present an approach aimed at enhancing robotic locomotion by leveraging the system's natural movements through nonlinear modes. Contrary to traditional kinematic control methods that impose predefined trajectories, often incompatible with the system's actual dynamics. Our strategy draws inspiration from biological systems that adapt their gait to their intrinsic dynamics. This allows us to avoid lockups and minimize energy consumption to the strict minimum required to compensate for friction. For example, a quadruped switches from trotting to galloping to increase its speed rather than trotting faster, thus aligning with its natural dynamics. Recent advances in structural dynamics enable the analysis of nonlinear modes without linearizing the system. When applied to robotics, these methods reveal that the optimal locomotion frequencies vary with the energy level and require multi-harmonic excitations. This approach offers prospects for defense applications, such as the development of assistance robots and transport drones. Our current work focuses on designing an underwater serpentine robot for karst exploration, inspired by the agility of eels in confined environments. By capitalizing on nonlinear modes still scarcely exploited in robotics, we aim to create systems that approach the efficiency of biological locomotion.
11:10pm-11:30pm, [Optional: discussion on forthcoming projects], break
11:30am-13:00am For those interested from IP-Paris, meeting in Gilles Kahn about the "future of CS at IP-Paris"
12:30am-1:30pm déjeuner - espace innovation
1:30pm-2:15pm Eric Goubault, Sylvie Putot (Ecole Polytechnique):, " A Zonotopic Dempster-Shafer Approach to the Quantitative Verification of Neural Networks ", video
Abstract: The reliability and usefulness of verification depend on the ability to represent the uncertainty appropriately. Most existing work on neural network verification relies on the hypothesis of either set-based or probabilistic information on the inputs. In this work, we rely on the framework of imprecise probabilities, specifically p-boxes, to propose a quantitative verification of ReLU neural networks, which can account for both probabilistic information and epistemic uncertainty on inputs. On classical benchmarks, including the ACAS Xu examples, we demonstrate that our approach improves the tradeoff between tightness and efficiency compared to related work on probabilistic network verification, while handling much more general classes of uncertainties on the inputs and providing fully guaranteed results.
2:15pm-3:00pm (All), Réunion restreinte avec Jean-Daniel Masson