**1. Introduction**

Unmanned vehicle technology and surface robots have been rapidly developed over the past few years. These systems supersede previously used methods for exploring the underwater parts of the Earth. Trends in the development of unmanned systems point clearly towards the future execution of underwater tasks, including hydrographical surveys, using the direct nearness to the bottom by autonomous surface vehicles (ASVs). The use of ASVs can supplement or replace many hours of measurements conducted by teams of hydrographers, especially in remote areas. Nowadays ASVs are used in many scientific and commercial implementations. They can be met for example in the army for reconnaissance and combat purposes, they serve as research units providing information on various aspects of the aquatic environment, and as carriers of measuring equipment for the inventory of watercourses and reservoirs.

The main tasks and challenges for ASVs depend on the kind of mission performed. However, some of them are common for all approaches and can be treated as a basis for specialized tasks. Among these for sure is the navigation itself and mission control with the use of telemetry, as the most basic priority for ship navigation is its safety. The navigation process can be however understood variably in different applications. In some approaches and applications, like simultaneous localization and mapping (SLAM), the term navigation also means getting information about the surrounding area. Various sensors for this purpose can be employed like 3D laser scanners [1]. These aspects, generally referred to as navigation, can also be found for example in [2,3], where lidar is also used for navigation

purposes. Some authors also include the path planning process into navigation itself, while the others threat dynamic path planning as more collision avoidance tasks. A fine survey on this can be found for example in [4,5].

One of the most important safety tasks during the operation of USV is avoiding collisions. It might be treated more as a situation awareness task and not navigation itself. Nevertheless the process of automating collision avoidance is always a key issue for unmanned vehicles, as it directly influences its safety and ability to perform a mission. Maneuvering for anti-collision purposes consists of several steps, including target detection, movement vector estimation, identifying the correct collision avoidance maneuver based on navigation obstacles and other moving ships, and finally implementation. The first step is always to get information about the surrounding environment, which is done by on-board sensors, processing the information for the anti-collision module. The most obvious sensors for this case of robotic application are range finders or in more advanced applications laser scanners [6]. The use of other sensors require implementation of advanced fusion algorithms, like in [7], where it is proposed to enhance the laser system with cameras. The information gathered from sensors is then processed by the anti-collision system to find the most suitable track. A fine survey on this, together with the review of sensors used for anti-collision in USV can be found for example in [8]. There are also some examples in literature for using radar sensor as a core for anti-collision, mostly in maritime ASV. Such approach can be found for example in [9] or [10] in which typical pulse X-band radar is used or in [11] in which frequency-modulated continuous-wave radar (FMCW) X-band radar is proposed (it might also be an option in [9]). These approaches are suitable for maritime application, however typical marine sensors may be too big for the smaller USV in inland waters. The new contribution of this research is to indicate a new approach, which is the implementation of autonomous radar for water vehicles.

In this paper, we propose a new idea, to implement an automotive 3D radar sensor in the autonomous navigation system of an ASV. The proposed approach is a combination of the traditional approach met in the waters, namely the use of radar sensors, and the approach known from roads in which radars are used for car collision systems. This sensor works with a fixed antenna, unlike the traditional rotated radar antennas. The first step in target tracking by radar is target detection, especially small targets in close range observation, which is essential in restricted waters.

In this study, experiments were conducted in real inland waterway conditions with an automotive 3D radar sensor mounted on an ASV owned by Marine Technology Ltd., named HydroDron. The goal was to use the collected data to address the autonomous collision avoidance problem for future intelligent ASV systems. This will be the basis for the implementation of such an innovative system for real-time ASV missions.

It should be pointed out here, that the sensor and system proposed in this study is suitable for small ASVs performing their duties in inland water or in restricted harbor areas. The detection range of this sensor is too small to be useful for marine vessels and therefore marine applications are beyond the scope of this paper. The use cases covered by this research assume that the ASV is performing her autonomic mission (likely hydrographic, but can be any other), navigating in a lake, river or near-coast waters. Hydrographic surveys are often performed in the areas near recreational or fishing sites. The goal of the research was to present detection possibilities for typical objects that could also be met in these areas, which includes not only boats, but also other floating objects.

The paper is organized as follows—in Section 2 the idea and theory of the radar used is presented; Section 3 gives the details of the anti-collision system concept together with the review of related works and papers; Section 4 provides a description of the research; and Section 5 includes the conclusions.

#### **2. Automotive Radar Sensors**

Automotive radar is used to detect objects in the vicinity of a vehicle. The sensor consists of a transmitter and receiver. The transmitter emits radio waves that return to the receiver after bouncing from the target. By controlling the direction in which radio waves are sent and received, it is possible to determine the distance, speed, and direction of the objects.

There are two basic methods for measuring distances using radar. The first is known as the direct propagation method, which measures the delay associated with receiving the reflected signal. The delay is correlated with the distance of the reflecting object depending on the speed of light and period and the transmission and reception of waves. The second method is known as the indirect propagation method. In the case of indirect propagation, a modulated frequency is sent and received, and the frequency difference can be used to directly determine the distance and relative velocity of the object. This requires controllable antennas that can be automatically routed or receive signals simultaneously from several different directions.

## *2.1. FMCW Radars*

There are two types of automotive radars, pulse radar or radar with continuous wave. The latter is termed the frequency-modulated continuous-wave radar (FMCW). The pulse radar sends short pulses and determines the distance by measuring the delay time between the transmitted and feedback signal [12]. The FMCW radar continuously sends a linearly modulated signal and determines the distance based on the difference in the transmitted and received frequencies, as shown in Figure 1.

**Figure 1.** Ranging with an FMCW system [13].

Measuring very short time periods in electronics is difficult, which means that building a good resolution pulse radar is very expensive. However, the resulting resolution is relatively precise, e.g., the FMCW radar can easily have a resolution of 0.5 m [14].

Impulse radars are blind at short distances—for example, the 50–100 m in front of the radar is usually a blind spot. FMCW radars do not have this problem. However, for long-range targets, the pulse radar is better due to the narrower bandwidth and less noise.

In both methods, if the target is moving, the motion creates a Doppler shift in the frequency of the transmitted radar waves. This is an additional advantage for the impulse radar because it can also measure the relative target speed. In the FMCW radar, this is a problem because the distance is measured by measuring the frequency difference between the transmitted and received radar waves, and each additional frequency offset caused by the Doppler effect of the moving target "shifts" the measured distance of the object. To remedy this problem, FMCW radar systems use several different modulation schemes, including modulation with increasing frequency and frequency reduction. If these offsets are not alleviated by algorithms or have very fast frequency changes, this effect may cause the appearance of ghost targets on the FMCW radar [15].

Automotive radars are divided into three categories—long-range radar (LRR), medium range radar (MRR), and short-range radar (SRR). LRR is used to measure the distance and speed of other vehicles, MRR is used in the wider field of view, and SRR is used to detect objects near the vehicle. Two main frequency bands are used in the automotive radar systems—24 GHz and 77 GHz [16,17].

SRR for vehicles uses 24 GHz frequency because the band can detect objects at short and medium distances. A 24 GHz radar is also used to detect an object that can be obstructed or is located very close. Radar systems with the same repeatability can also be used to detect dead spots, which directly involves avoiding collisions. SRR sensors are not used to measure the angle of the detected objects and have a very wide side coverage. Usually, they are operated in pulse mode or in continuous wave mode. Small range radars require a controllable antenna with a large scan angle, creating a wide field of view [18].

While difficult to implement, LRR uses the higher permitted transmission power (77 GHz) to obtain better performance. It is easier to develop 24 GHz bands, but more difficult to integrate such radars systems into the vehicle due to their larger size. In addition, these sensors work with the same performance as the 77 GHz radars but with three times larger antennas. Therefore, the 77 GHz radar is smaller and, in contrast to the 24 GHz radar, is easier to integrate with a vehicle at a lesser cost. An additional and undeniable advantage of the wider 77 GHz band is that it provides drivers with a better resolution of objects by providing greater accuracy. The detection and reaction to the presence of both large and small objects are possible due to the clever signal processing. In the case of LRR, a higher resolution is also provided for a more limited scanning range, which requires a larger number of directional antennas [16,17].

In the 77–81 GHz range, bandwidths up to 4 GHz are available, while the bandwidth available in the 24 GHz band is 200 MHz. The difference between the frequency of the signal emitted by the transmitter and the frequency of the received reflected signal is linearly dependent on the distance from the transmitter to the object. The accuracy of measuring this distance and resolution are important. The resolution is understood as the minimum distance between the objects so that they can be distinguished as different. The transition from 24 GHz to 77 GHz results in a 20 times better performance in terms of resolution and accuracy. The resolution of the 77 GHz radar range can reach 4 cm. For comparison, the 24 GHz radar achieves a resolution of 75 cm. Therefore, the advantage of the 77 GHz system is that it can detect objects that are at a short distance away from each other. Finally, the wavelength of 77 GHz signals is a third the frequency of the 24 GHz system, which enables significantly smaller modules in the spatially limited areas of the vehicle. The relative antenna sizes are shown in Figure 2.

**Figure 2.** Antenna sizes for 24 GHz (left) and 77 GHz (right) radar systems [19].

Figure 3 shows the range and width of coverage of SRR, MRR, and LRR. LRR can detect objects in a wide area and can cover a range from 10 to 150 m at a beam width of 10◦. MRR can cover a range up to 50 m with a beam width of 30◦. In contrast, SRR can be used to track objects within a distance of 20 m from the vehicle with a beam width of 60◦ [20–22].

**Figure 3.** The range and width of coverage of short-range radar, medium-range radar and longrange radar [19].
