| Summary: | As recent years have seen an increase in use of robotics in both public and private spaces, evaluations to ensure safe and efficient navigation through common scenarios are needed. The same evaluations can improve simulations of human behaviour used in for example evacuation simulations. There exist many different collision avoidance algorithms to aid the navigation, all of which can display problematic behaviours under different circumstances. This paper evaluates Reciprocal Obstacle Avoidance (RVO) and Optimal Reciprocal Collision Avoidance (ORCA), two popular approaches to collision avoidance based on Velocity Obstacles. They are evaluated in a total of 5 scenarios put together with elements from common real life scenarios. This includes groups of agents interacting with each other in head-on meeting and crossing configurations, space constrictions in the form of static walls to form so called open and corridor space scenarios, as well as a narrow doorway. Each scenario was run with n agents where n ∈ {2, 4, 6, 8, 10, 12, 14, 16, 18, 20}. It was found that increasing agent numbers in corridor spaces introduces a high risk of so called abductions, where some agents are forced to take extreme detours. It was also found that the number of collisions followed the number of agents in a simulation and was especially high for corridor scenarios. Overall, the thesis highlights factors that might cause problematic behaviours for agents navigating in real and simulated common spaces and provides insights that can light the way in future work aiming to improve such navigation. Mängden robotik i både offentliga och privata utrymmen har ökat under de senaste åren. Därför behövs utvärderingar för att säkerställa säker och effektiv navigering i vardagliga scenarier. Dessa utvärderingar kan även bidra till en förbättring av simuleringar av mänskligt beteende som används i exempelvis krissimuleringar. Det finns många algoritmer som syftar till att undvika kollisioner under pågående navigering. Olika algoritmer riskerar ...
|
|---|