Thierry Fraichard

Growing Hidden Markov Models: An Incremental Tool for Learning and Predicting Human and Vehicle Motion

Modeling and predicting human and vehicle motion is an active research domain. Owing to the difficulty in modeling the various factors that determine motion (e.g. internal state, perception) this is often tackled by applying machine learning techniques to build a statistical model, using as input a collection of trajectories gathered through a sensor (e.g. camera, laser scanner), and then using that model to predict further motion. Unfortunately, most current techniques use offline learning algorithms, meaning that they are not able to learn new motion patterns once the learning stage has finished. In this paper, we present an approach where motion patterns can be learned incrementally, and in parallel with prediction. Our work is based on a novel extension to hidden Markov models, called growing hidden Markov models, which gives us the ability to learn incrementally both the parameters and the structure of the model. The proposed approach has been evaluated using synthetic and real trajectory data. In our experiments our approach consistently learned motion models that were more compact and accurate than those produced by two other state-of-the-art techniques.

Inevitable Collision States: A probabilistic perspective

For its own safety, a robot system should never find itself in a state where there is no feasible trajectory to avoid collision with an obstacle. Such a state is an Inevitable Collision State (ICS). The ICS concept is particularly useful for navigation in dynamic environments because it takes into account the future behaviour of the moving objects. Accordingly it requires a model of the future evolution of the environment. In the real-world, the future trajectories of the obstacles are generally unknown and only estimates are available. This paper introduces a probabilistic formulation of the ICS concept which incorporates uncertainty in the model of the future trajectories of the obstacles. It also presents two novel probabilistic ICS-checking algorithms that are compared with their deterministic counterpart.

Provably safe navigation for mobile robots with limited field-of-views in dynamic environments

This paper addresses the problem of navigating in a provably safe manner a mobile robot with a limited field-of-view placed in a unknown dynamic environment. In such a situation, absolute motion safety (in the sense that no collision will ever take place whatever happens in the environment) is impossible to guarantee in general. It is therefore settled for a weaker level of motion safety dubbed passive motion safety: it guarantees that, if a collision takes place, the robot will be at rest.The primary contribution of this paper is the concept of Braking Inevitable Collision States (ICS), i.e. a version of the ICS corresponding to passive motion safety. Braking ICS are defined as states such that, whatever the future braking trajectory followed by the robot, a collision occurs before it is at rest. Passive motion safety is obtained by avoiding Braking ICS at all times.It is shown that Braking ICS verify properties that allow the design of an efficient Braking ICS-Checking algorithm, i.e. an algorithm that determines whether a given state is a Braking ICS or not.To validate the Braking ICS concept and demonstrate its usefulness, the Braking ICS-Checking algorithm is integrated in a reactive navigation scheme called PassAvoid. It is formally established that PassAvoid is provably passively safe in the sense that it is guaranteed that the robot will always stay away from Braking ICS no matter what happens in the environment.

Collision avoidance in dynamic environments: An ICS-based solution and its comparative evaluation

This paper presents ICS-AVOID, a collision avoidance scheme based upon the concept of Inevitable Collision State (ICS), ie a state for which, no matter what the future trajectory of the robotic system is, a collision eventually occurs. By design, ICS-AVOID can handle dynamic environments since ICS do take into account the future behaviour of moving objects. ICS-AVOID is designed to keep the system away from ICS. By doing so, motion safety is guaranteed (by definition a robotic system in a non-ICS state has at least one collision-free trajectory that it can use). To demonstrate the efficiency of ICS-AVOID, it has been extensively compared with two state-of-the-art collision avoidance schemes: the first one is built upon the Dynamic Window approach and the second one on the Velocity Obstacle concept. The results obtained show that, when provided with the same amount of information about the future evolution of the environment, ICS-AVOID outperforms the other two schemes. The first reason for this has to do with the extent to which each collision avoidance scheme reasons about the future. The second reason has to do with the ability of each collision avoidance scheme to find a safe control if one exists. ICS-AVOID is the only one which is complete in this respect thanks to the concept of Safe Control Kernel.

An anthropomorphic navigation scheme for dynamic scenarios

This paper is concerned with the navigation of personal robots in human-populated environments. The behavior of a person among its peers is governed by a number of unspoken social rules, e.g. maintaining an appropriate distance. The primary contribution of this paper is a navigation scheme that is anthropomorphic, i.e. that emulates human behaviors and seeks to adhere to these social rules. Unlike previous works in this area, the focus herein is on dynamic scenarios. The navigation scheme proposed explicitly reasons on the future behavior of the people involved so as to produce better socially acceptable trajectories (not to mention safer trajectories as well). The navigation scheme relies upon a novel cost function called the social costmap that captures in a unified way the different social rules imposed by the people populating the robots workspace.

Motion estimation from range images in dynamic outdoor scenes

Object-class independent motion estimation from range data is a challenging task. We present here a novel approach that is able to derive a dense motion field based on range images only. We propose to first segment the range image into segments using a recently proposed segmentation criterion. Motion is then estimated segment-wise with full 6 degrees of freedom. To that end, we introduce dynamic mapping, i.e. the accumulation of measurements for moving objects. We show experimentally that the approach is able to deliver a dense motion field which can then be used for object-class independent trajectory estimation.

An efficient and generic 2D Inevitable Collision State-checker

An inevitable collision state (ICS) for a robotic system is a state for which, no matter what the future trajectory of the system is, a collision eventually occurs. ICS can be used for both motion planning (to reduce the search space) and reactive navigation (for obvious safety reasons, a robotic system should never ever move to an ICS). ICS are particularly suited for navigation in dynamic environments since they take into account the future behaviour of the moving objects. Using ICS in practice is difficult given the intrinsic complexity of their characterization. The main contribution of this paper is a generic and efficient ICS-Checker, ie an algorithm that determines whether a given state is an ICS or not, for planar robotic systems with arbitrary dynamics moving in dynamic environments. The efficiency is obtained by applying the following principles: (a) reasoning on 2D slices of the state space of the robotic system, (b) precomputing off-line as many things as possible, and (c) exploiting graphics hardware performances. The ICS-Checker has been applied to two different robotic systems: a car-like vehicle and a spaceship. It has also been integrated in a reactive navigation scheme to safely drive the car-like vehicle.

Navigating dynamic environments using trajectory deformation

Path deformation is a technique that was introduced to generate robot motion wherein a path, that has been computed beforehand, is continuously deformed on-line in response to unforeseen obstacles. In an effort to improve path deformation, this paper presents a trajectory deformation scheme. The main idea is that by incorporating the time dimension and hence information on the obstaclespsila future behaviour, quite a number of situations where path deformation would fail can be handled. The trajectory represented as a space-time curve is subject to deformation forces both external (to avoid collision with the obstacles) and internal (to maintain trajectory feasibility and connectivity). The trajectory deformation scheme has been tested successfully on a planar robot with double integrator dynamics and a car-like vehicle.

From path to trajectory deformation

Path deformation is a technique that was introduced to generate robot motion wherein a path, that has been computed beforehand, is continuously deformed on-line in response to unforeseen obstacles. This paper introduces the first trajectory deformation scheme as an effort to improve path deformation. The main idea is that by incorporating the time dimension and hence information on the obstacles future behaviour, quite a number of situations where path deformation would fail can be handled. The trajectory deformation scheme presented operates in two steps, ie, a collision avoidance step and a connectivity maintenance step, hence its name 2-step-trajectory-deformer (2-STD). In the collision avoidance step, repulsive forces generated by the obstacles deform the trajectory so that it remains collision-free. The purpose of the connectivity maintenance step is to ensure that the deformed trajectory remains feasible, ie, that it satisfies the robots kinematic and/or dynamic constraints. Moreover, unlike path deformation wherein spatial deformation only takes place, 2-STD features both spatial and temporal deformation. It has been tested successfully on a planar robot with double integrator dynamics moving in dynamic environments.

Real-time trajectory generation for car-like vehicles navigating dynamic environments

This paper presents Tiji, a trajectory generation scheme, ie an algorithm that computes a feasible trajectory between a start and a goal state, for a given robotic system. Tiji is geared towards complex dynamic systems subject to differential constraints, such as wheeled vehicles, and its efficiency warrants it can be used in real-time. Above all, Tiji is able to compute a trajectory that reaches the goal state at a prescribed final time in order to avoid collision with the moving objects of the environment. The method proposed, which relies upon a parametric trajectory representation, is variational in nature. The trajectory parameters are incrementally updated in order to optimize of a cost function involving the distance between the end of the trajectory computed and the (goal state, final time) pair. Should the goal state be unreachable (if the final time is ill-chosen), the method returns a trajectory that ends as close as possible to the (goal state, final time) pair, which can be useful in certain applications.