Rahul Kala

In this paper, we propose a Voronoi Heuristic-Assisted Multi-Objective Optimization (VH-MOO) framework for motion planning in complex environments. Optimization techniques have the potential to give a feasible path in a complex environment; however, the approaches need to be near-real time to adjust to dynamically changing environments. In this paper, we first consider a multi-objective optimization technique to simultaneously consider multiple conflicting objectives, such as minimizing travel time, increasing clearance, and reducing path length, and further combine the optimization with the Voronoi diagram's ability to identify high-clearance regions for computational efficiency. By incorporating Voronoi heuristics, the framework efficiently narrows down feasible regions, allowing for faster and more reliable motion planning solutions. Experimental results demonstrate the effectiveness of our method compared to traditional motion planning techniques, highlighting its potential for real-time applications in robotics and autonomous systems.

This paper proposes an eco-driving technique for an ego vehicle operating behind a non-communicating leading Heavy-Duty Vehicle (HDV), aimed at minimizing energy consumption while ensuring inter-vehicle distance.Anovel data-driven approach based on Deep Reinforcement Learning (DRL) is developed to predict the future speed trajectory of the leadingHDVusing simulated speed profiles and road slope information. The DQN-based speed predictor achieves a prediction accuracy of 95.4% and 93.2% in Driving Cycles 1 and 2, respectively. This predicted speed is then used to optimize the ego vehicle’s speed plan through a distributionally robust Model Predictive Controller (MPC), which accounts for uncertainties in the prediction, ensuring operational safety. The proposed method demonstrates energy savings of 12.5% in Driving Cycle 1 and 8.6% in Driving Cycle 2, compared to traditional leading vehicle speed prediction methods. Validated through case studies across simulated and real-world driving cycles, the solution is scalable, computationally efficient, and suitable for real-time applications in Intelligent Transportation Systems (ITS), making it a viable approach for enhancing sustainability in non-communicating vehicle environments.

Future warehouses will see robots autonomously picking up all the items ordered by the user and delivering them for packaging, sorting and delivery. This requires warehouses to schedule the robots to autonomously cater to all customer orders. In this paper, we solve the problem of batching orders, where several user orders are clubbed into one to be collectively picked by one robot, scheduling the visit of the robot to pick up the desired items and assigning one robot to carry the operation. Each robot is assumed to have a small payload capacity that restricts the number of orders that can be batched. Each item is assumed to be found at several locations, and the robot can select anyone of the locations. The proposed solution uses heuristics to create initial batches that are further optimised by using Genetic Algorithm. Genetic Algorithm also jointly optimises the batching and item pick sequencing. The selection of one out of several sites is done by using a Dynamic Programming approach. The solution is post processed by using a Dynamic Programming solution for optimally scheduling the robot visits within every batch. The experimental results show that the proposed solution scales to a large number of orders because of the use of heuristics, while the results are better than the use of greedy heuristic approaches.

Localization is one of the key aspect of any automated navigation system, it means finding own position in a given environment, without it an agent might get lost. In a dynamic environment localization is a very difficult task because various factors such as weather and illumination may change over time. We developed a low-cost solution to the localization problem, relying only on a camera sensor, without using expensive sensors like lidars. The solution allows to handle dynamic nature of the environment and uses previous knowledge for localization, hence it is named as Experience Based Localization. STREET-CNN is a core part of the method which is trained on STREET dataset. This dataset contains several images of streets in IIIT-A along with their GPS coordinates as labels. The trained network calculates the similarity score between the two cross domain images. Presented method is equipped with Kalman Filter to include speed variations of agent while predicting its location. Results produced with domain adaptation technique is compared with, without the domain adaptation technique.