We present a new traffic dataset, Meteor, which captures traffic patterns and multi-agent driving behaviors in unstructured scenarios. Meteor consists of more than 1000 one-minute videos, over 2 million annotated frames with bounding boxes and GPS trajectories for 16 unique agent categories, and more than 13 million bounding boxes for traffic agents. Meteor is a dataset for rare and interesting, multi-agent driving behaviors that are grouped into traffic violations, atypical interactions, and diverse scenarios. Every video in Meteor is tagged using a diverse range of factors corresponding to weather, time of the day, road conditions, and traffic density. We use Meteor to benchmark perception methods for object detection and multi-agent behavior prediction. Our key finding is that state-of-the-art models for object detection and behavior prediction, which otherwise succeed on existing datasets such as Waymo, fail on the Meteor dataset. Meteor is a step towards developing more sophisticated perception models for dense, heterogeneous, and unstructured scenarios.

Visual Simultaneous Localization and Mapping (VSLAM) tend to face challenges with low-framerate and low-resolution data, as well as in congested environments with moving objects and occlusions, leading to increased drift. To enhance the robustness of VSLAM, semantics are increasingly being employed to address specific limitations. This work adopts a comprehensive approach, incorporating sparse features from semantics for indoor environments. During training, a prior semantic map is built, while some known places relative to the semantic map are recorded, which are then stored in a database. An active place recognition module matches semantics with those in the database. Additionally, a hybrid optimization module estimates the robot’s pose by minimizing semantic reprojection errors, ensuring pose proximity to detected places, preserving semantic size consistency, incorporating live feature maps, and ensuring the pose to be within navigable space using an occupancy map made a priori. Experimental comparison demonstrates superior performance over conventional VSLAM, yielding lower error values.

The stereo correspondence problem is one of the most pre-eminent problems in a stereo vision system. With the right correspondence, a stereo vision system can help cap over diverse problems, while on the other hand, a wrong correspondence can be costly. While the performance of a feature-based correspondence approach is exceptional, the method can still produce wrong correspondences. This work presents an amalgam of feature-based and correlation-based correspondence, where the local pixels around a feature pair are compared using Structural SIMilarity index (SSIM), enhancing the correspondences, and a semantic-based filtering module, which further filters the obtained corresponding features using semantic data whenever detected in both the stereo image pair. While approaches in the literature are focused towards finding better features and their representation, the proposed approach advocates that correlation-based verification of the features can filter out bad correspondences, and in addition, aided by semantic-level filtering. These two modules establish the novelty of the work. The proposed correspondence matching algorithm is used to solve the problem of Visual Odometry to let a low-cost robot compute its pose in a novel environment. The experimental results show adequate filtering of wrong feature correspondence wherein, different environments with different lighting conditions were also considered. The proposed approach outperformed numerous state-of-the-art approaches available in the literature. The visual odometry algorithm using the proposed correspondence matching is compared against classical methods and a deep learning method, and it is observed that the proposed approach delivers lower trajectory error values in most scenarios on the KITTI dataset sequences.

Robot mission planning deals with accomplishing a mission that requires the robot to visit a set of sites and carrying out specific tasks at each site, consisting of Boolean and temporal constraints. The Boolean constraints are like "Visit any one of the three coffee machines" and "Visit any two of the three instructors", while the temporal constraints typically ask the robot to carry some sub-missions in a sequential order and some others in parallel. Each mission can have a different number of sites and nature of constraints involved. Each mission is expressed in a generalized form using OR, AND, and THEN clauses. Further, the notion of sensors is added, wherein a resource for a mission may be unavailable and therefore the robot will have to plan an alternate (maybe longer) strategy, however the availability will only be known when the robot physically visits the mission site. The typical mechanisms to solve for such mission using formal languages and verification engines incur an exponential complexity. Therefore, this paper proposes an evolutionary algorithm to solve the problem, enabling the use of probabilistic optimality properties. However, calculating the expected path length with the prior probabilities of the sensor is also exponential complexity which causes the resulting algorithm to have an exponential complexity in terms of the number of sensors. A greedy heuristic cost function is devised that imitates human decision making, wherein the robot tries to solve the mission with the shortest length as well as maximize the probability of completion of the task. Further, each site is assumed to have an associated waiting time which is the expected time for which the robot will have to wait in the queue till the resource is available for use. Another greedy heuristic is used to align the sites in the crossover operation to approach a better solution faster. Experimental results confirm that the proposed solution performs well as compared to baseline optimistic and pessimistic approaches. The proposed algorithm is also tested on the Pioneer LX robot.

The problem of mission planning is to enable robots solve complex missions with Boolean and temporal operators in the mission specification. Typically, missions are specified using a Linear Temporal Logic Formulation and solved by using a model verification approach which has an exponential complexity. Given a language which is polynomial verifiable, evolutionary paradigm of mission planning can enable probabilistic optimality and probabilistic completeness, thus enabling the use of solvers for a very high number of variables, which is impossible to do using the model verification techniques. This paper motivates the heuristic of a mission consisting of a number of tasks, such that each task is a complex instruction given by a user, while many such users share a robot. The heuristic is used to generate a near-optimal solution of tasks that can then be fused optimally by a Dynamic Programming approach to make the solution of the mission. However, the problem is not decomposable and optimal solutions of tasks do not result in an optimal solution to mission. Hence a 2step algorithm is used. The first step computes a near-optimal solution of the tasks. The second step does a full Genetic Algorithm search to generate task solutions that eventually fused by a Dynamic Programming approach produce an optimal mission solution. Comparative analysis is done with numerous baselines and the proposed approach is experimentally shown to perform better than all baselines. The experiments are also done on the Pioneer LX robot using the Robot Operating System framework.

Abstract Narrow corridors are a cause of intense problems in sampling‐based approaches due to the small probability of generation of samples inside the narrow corridor. The obstacle‐based sampling and bridge test sampling techniques rely on generating fresh samples at every iteration. Memory of the forbidden configuration space can lead to the discovery of new narrow corridors or generating additional key samples inside the same narrow corridor. Hence, in this paper, it is proposed to additionally solve the problem of generating a roadmap in the forbidden configuration space, called as the dual roadmap. The dual roadmap, so generated, has vertices that are collision‐prone and stores the structure of the forbidden configuration space. To reduce memory and computation time, only the boundary of the forbidden configuration space is stored, which is more informative. The dual roadmap, so constructed, is used to generate valid samples inside the middle of the narrow corridors. The construction of the additional roadmap takes a very small time and memory and is largely the biproduct of obstacle‐based sampling that is normally thrown away. Experimental results show that the proposed sampling method is very effective for finding narrow corridors as compared to popular sampling methodologies existing in the literature.

The classic problem of robot motion planning asks the robot to go from A to B avoiding obstacles. Missions are challenging problems asking the robot to visit a set of sites to accomplish a mission. The mission planning problems are largely studied as a Travelling Salesman Problem involving combinatorial optimization. In this paper the problem is generalized to any Boolean expression, giving more expressing powers to specify missions like “Visit any one of three coffee machines” or “Visit any two of three instructors”, along with other mission sites to be mandatorily visited. The problem is solved using multiple robots in a decentralized manner. The Boolean expression is simplified into an `OR of AND' format, which gives the flexibility to solve all the AND components and to select the minimum cost solution among them. Each of the AND components is a reduced multi-robot Travelling Salesman Problem solved by using k-medoids clustering and evolutionary computation. The results obtained by this approach are compared with the centralized algorithm and a master slave algorithm which uses a randomized algorithm for robot assignment, and for every such assignment the corresponding optimization problem of visiting the sites is solved for. The comparison depicts that as the problem size and the number of robots increase, the decentralized approach outperforms the rest enormously. The results are also tested on a Pioneer LX robot working in an office environment to carry dummy missions of everyday needs.

The problem of mission planning enables a robot to solve for complex missions and thereafter execute the missions. The problem is seen as an advancement of the classical problem of motion planning wherein the task is to find a trajectory for a robot to go from a configuration A to a configuration B avoiding obstacles. The popular mechanism to solve the problem is using Temporal Logic specifications to specify a mission, and to thereafter use search strategies to solve the mission. The same requires an exponential complexity in terms of propositional variables to search and verify the mission plan. Further, optimality is usually not a criterion in finding a solution, and the transition costs are usually ignored leading to sub-optimal mission plans. As missions get more complex using a large number of propositional variables for specification, it may no longer be possible to use exponential complexity algorithms. The paper proposes the use of Evolutionary Computation to solve the same problem. We first develop a new restricted language for mission design. Even though the language is restrictive, it can specify a large number of missions of real life service robotics. Then we design an evolutionary computation framework to solve the mission. Probabilistic Roadmap technique is used to get the transition system and transition costs between regions of interest. The mission planner takes the mission specification and these transition costs to compute a mission plan. The mission plan is executed using a reactive navigator that can avoid any dynamic obstacle, other people and robots.

Robots of the future would be sophisticated enough to do all kinds of tasks for the human master, while operating a team of robots would ensure that the tasks are done as efficiently as possible. This paper presents a solution to a very common mission wherein a team of robots needs to visit a number of mission sights and carry some operation there. The sights may have their own preference of robots. The real world workspace is first converted into a roadmap using the Probabilistic Roadmap algorithm. The adopted technique for roadmap generation uses a hybrid of narrow corridor sampler, obstacle based sampler and uniform sampler; with hybrid edge connection technique that aims to first connect disconnected roadmaps and then introduces redundant cycles and also ensures a good coverage of the configuration space. The roadmap is used to compute a cost matrix between every mission sight. This cost matrix is used by an optimization algorithm which deputes different mission sights to different robots as well as dictates the order of visit of the missions by each robot. Local optimization is used to quickly compute the optimal plan. The navigation of the robots is done using a Fuzzy Logic based navigator. Experiments show that the robots are able to coordinate with each other and complete the mission as a team very effectively.

Coordination in the problem of multirobot path planning enables the individual robots to escape collision when planned in a decentralized manner. A good coordination strategy should not be time consuming, should be probabilistically complete, and should not overly prefer a robot. In this article, an altering local search-based strategy is studied. Experiments reveal the ability of the algorithm to place the different robots judiciously for near-optimal collision avoidance. The algorithm takes less time compared to the centralized approaches, is probabilistically complete as opposed to the reactive approaches, and treats the robots alike as opposed to the prioritized approaches.

A bulk of research is being done for the autonomous navigation of a mobile robot. Multirobot motion planning techniques often assume a direct communication among the robots, which makes them practically unusable. Similarly, approaches assuming the robot moving amid humans assume cooperation of humans, which may not be the case if the human is replaced by a robot. In this paper, a deliberative planning at the higher level with a new cell decomposition technique is presented, along with a reactive planning technique at the finer level, which uses fuzzy logic. Coordination among the robots in the absence of direct communication and knowledge of other robot's intent is a complex research question, which is solved using a simple fuzzy-based modeling. Experimental results show that the multiple robots maintain comfortable distances from the obstacles, navigate by near optimal paths, can easily escape previously unseen obstacles, and coordinate with each other to avoid collision as well as maintain a large separation. This work displays a simple and easy to interpret system for solving complex coordination problem in multirobotics.

Planning of autonomous vehicles in the absence of speed lanes is a less-researched problem. However, it is an important step toward extending the possibility of autonomous vehicles to countries where speed lanes are not followed. The advantages of having nonlane-oriented traffic include larger traffic bandwidth and more overtaking, which are features that are highlighted when vehicles vary in terms of speed and size. In the most general case, the road would be filled with a complex grid of static obstacles and vehicles of varying speeds. The optimal travel plan consists of a set of maneuvers that enables a vehicle to avoid obstacles and to overtake vehicles in an optimal manner and, in turn, enable other vehicles to overtake. The desired characteristics of this planning scenario include near completeness and near optimality in real time with an unstructured environment, with vehicles essentially displaying a high degree of cooperation and enabling every possible (safe) overtaking procedure to be completed as soon as possible. Challenges addressed in this paper include a (fast) method for initial path generation using an elastic strip, (re-)defining the notion of completeness specific to the problem, and inducing the notion of cooperation in the elastic strip. Using this approach, vehicular behaviors of overtaking, cooperation, vehicle following, obstacle avoidance, etc., are demonstrated.

Chaotic traffic, prevalent in many countries, is marked by a large number of vehicles driving with different speeds without following any predefined speed lanes. Such traffic rules out using any planning algorithm for these vehicles which is based upon the maintenance of speed lanes and lane changes. The absence of speed lanes may imply more bandwidth and easier overtaking in cases where vehicles vary considerably in both their size and speed. Inspired by the performance of artificial potential fields in the planning of mobile robots, we propose here lateral potentials as measures to enable vehicles to decide about their lateral positions on the road. Each vehicle is subjected to a potential from obstacles and vehicles in front, road boundaries, obstacles and vehicles to the side and higher speed vehicles to the rear. All these potentials are lateral and only govern steering the vehicle. A speed control mechanism is also used for longitudinal control of vehicle. The proposed system is shown to perform well for obstacle avoidance, vehicle following and overtaking behaviors.

In this article, we propose a new algorithm to solve the problem of robotic path planning in static environment where the source and destination are given. A grid-based map has been used to represent the robotic world. The basic algorithm is built on an evolutionary approach, where the path evolves along with generations with each generation adding to the maximum possible complexity of the path. Along with complexity we optimise the total path length as well as the minimum distance from the obstacle in the robotic path. It may be seen that the value of evolutionary parameter number of individuals as well as the maximum complexity is less at start and more at the later stages of the algorithm. We use a Gaussian increase in these values whose parameter may be adjusted to control the time and output. Seven genetic operators have been implemented that include selection, crossover, soft mutation, hard mutation, insert, delete and elite. The phenotype representation consists of the coordinate where the robot is supposed to make a turn. This happens by the traversal of the path using these points by the evolutionary algorithm. Momentum determines the speed of the algorithm in this traversal.

The problem of path planning deals with the computation of an optimal path of the robot, from source to destination, such that it does not collide with any obstacle on its path. In this article we solve the problem of path planning separately in two hierarchies. The coarser hierarchy finds the path in a static environment consisting of the entire robotic map. The resolution of the map is reduced for computational speedup. The finer hierarchy takes a section of the map and computes the path for both static and dynamic environments. Both the hierarchies make use of an evolutionary algorithm for planning. Both these hierarchies optimize as the robot travels in the map. The static environment path is increasingly optimized along with generations. Hence, an extra setup cost is not required like other evolutionary approaches. The finer hierarchy makes the robot easily escape from the moving obstacle, almost following the path shown by the coarser hierarchy. This hierarchy extrapolates the movements of the various objects by assuming them to be moving with same speed and direction. Experimentation was done in a variety of scenarios with static and mobile obstacles. In all cases the robot could optimally reach the goal. Further, the robot was able to escape from the sudden occurrence of obstacles.