Skip to main content

AWoid (Collision Avoidance)

AWoid is a project carried out by ELIF LAB srl (
The main goal of the AWoid project is developing an artificial intelligence solution that could allow the onboard camera input of an autonomous vehicle, drone or robot to be used to estimate the probability of collision with obstacles during the navigation and to autonomously activate the right indications (action and steering angle) in order to avoid these obstacles and continue the navigation safely.


Technical Category
Robotics and automation

What is the challenge that is being addressed?

When autonomous vehicles, drones and robots navigate within outdoor and indoor spaces, it is important that they are able to perform their tasks safely, avoiding obstacles in the path or stopping if they risk harming people or property. Information from various on-board sensors can be used to perform these tasks. Among these, a relatively low-cost component is represented by the images collected by on-board cameras, which can provide valuable information to better direct the activities of the robot or autonomous vehicle. The aim, also in order not to depend on external connections and to preserve privacy, is to process these images directly on the device.

What is the AI solution the project plans to implement?

AWoid is a hybrid artificial intelligence solution: it combines neural networks with a semantic approach. On the one hand, this makes it possible to carry out training starting from data; on the other hand, the system is able to accept and incorporate an a priori competence introduced by the user through abstract rules and to produce in turn abstract commands readable by a human being.
The advantages of this type of approach are several:

  • Ability to work with incomplete datasets by introducing a priori information and human knowledge on the problem to be solved, potentially bypassing an expensive complete training process. On the other hand, if a complete training set is available, it is possible to train the developed structures (including semantic and dynamic graphs) integrating new knowledge through alternative processes.
  • Direct experience + external competence: AWoid represents a compromise between Neural Networks (that can learn from examples but are not directly modifiable) and logic rules (modifiable but cannot learn from examples). We see this process as similar to the learning process of a child, learning from direct experience but also from suggestions given by caring adults.
  • Transparency: process is readable, it is possible to intervene to evaluate it in all its steps. A plus also from an ethical point of view, as safety can be seen as the ability to predict or modify the actions proposed by an AI system.
  • Modularity: the system is modular and the different components are pluggable. Improvements can be applied on each of the individual blocks, not least because the process and output of each is exposed in the logs produced by the solution.

AWoid can take into account the specific characteristics of the vehicle and propose the best next action when encountering an obstacle (stop, steering left, steering right + steering angle).
The solution can be adapted to different architectures: it proved to be accurate and fast enough to work on x86_64 and Raspberry Pi4b. 

How will BonsAPPs support you in implementing this solution?

BonsAPPs offered economic, technical and business support for the development of our solution, access to a community of AI Talents and the promotion of AWoid’s core concepts to a wider market. On the technological side, the BonsApps Marketplace tools have enabled us to develop faster on multiple target devices. The Marketplace will also be an important showcase for us to reach out to new stakeholders and AI Talents and establish new collaborations.

bonsapps logo