Performance evaluation of a radar-on-chip system for obstacle detection strategies for urban air mobility (UAM) applications Open Access

AAM

Advanced Air Mobility

ADC

Analog-to-Digital Conversion

AoA

Angle of Arrival

ASSURED UAM

Acceptance Safety and Sustainability Recommendations for Efficient Deployment of UAM (European project)

BIST

Built-In Self-Test

BVLOS

Beyond Visual Line of Sight

CAN

Controller Area Network

CIRA

Italian Aerospace Research Centre

CTR

Controlled Traffic Region

DAA

Detect and Avoid

DEP

Distributed Electric Propulsion

DSP

Digital Signal Processor

DSS

Digital Signal Subsystem

FFT

Fast Fourier Transform

FMCW

Frequency Modulated Continuous Wave

GCS

Ground Control Station

GNSS

Global Navigation Satellite System

GPS

Global Positioning System

INS

Inertial Navigation System

JTAG

Joint Test Action Group

LNA

Low-Noise Amplifiers

MIMO

Multiple Input Multiple Output

ML

Machine Learning

MSS

Master Subsystem

PA

Power Amplifier

RCS

Radar Cross-Section

SBC

Single Board Computer

SNR

Signal-to-Noise Ratio

UAM

Urban Air Mobility

UART

Universal Asynchronous Receiver-Transmitter

UAS

Unmanned Aircraft System

UAV

Unmanned Aerial Vehicles

USB

Universal Serial Bus

VLL

Very Low Level

VTOL

Vertical Take-off and Landing

VLOS

Visual Line of Sight

Urban air mobility (UAM) has been proposed as potential solution for urban transport challenges, for both goods delivery and passenger transport. This new air transportation modality could be effective in reducing traffic congestion during daily commutes around cities while also promoting transport decarbonization and more sustainable mobility due to the use of exclusively electric vehicles. UAM is defined as a new concept based on the use of vertical take-off and landing (VTOL) vehicles for air transportation of people and goods in urban and suburban areas (ASSURED UAM project team, 2021; Ariante and Del Core, 2025). Indeed, UAM aims to adopt DEP to be mounted on VTOL aircraft that will be able to carry a few passengers, from 2 up to 7, reducing CO₂ emission and noise level (Menichino et al., 2022; Di Vito et al., 2023).

However, existing technologies and regulations only allow the implementation of the UAM concept with conventional helicopters. Nevertheless, based on publications and recent projects of the last decade, UAM is expected to be the future of the metropolitan urban mobility (Almaita et al., 2019). The development phase for UAM vehicles is well underway, with numerous companies having successfully built and flight-tested full-scale prototypes over the past decade (ASSURED UAM project team, 2021). Most UAM prototypes are designed with more than four rotors and propellers, utilize electric propulsion systems and can carry between two and five passengers. These prototypes exhibit flight characteristics more like helicopters than fixed-wing aircraft, operating at relatively low altitudes and in proximity to buildings. The general public is accustomed to the safety standards associated with large transport aircraft and regional fixed-wing aircraft used in commercial air transport (Goyal et al., 2018; Hasan, 2019).

UAM vehicles will have the best chances for full-scale implementation, provided that they are focused on safety, smart operations and connected under the supervision of a centralized platform. Safety, of course, always needs to be the priority, so any UAM vehicle needs to be outfitted with power redundancy and backup systems. Furthermore, UAM vehicles have to be conceived as “smart” vehicles with high autonomy level, which not only obviates the need for an onboard pilot and the associated costs but also enhances safety and makes the vehicle more controllable (Cohen et al., 2021; Ariante and Papa, 2024). An important role in the integration of UAM in the city context will be played by obstacle detection and avoidance systems. Thanks to the presence of those technologies, drones and unmanned aerial vehicles (UAVs) in general will be able to integrate perfectly into urban space, avoiding collisions with other aircraft, buildings, people and other obstacles (Menichino et al., 2023a, b). These systems, therefore, guarantee the safety of remote piloting, both under visual line of sight (VLOS) and beyond line of sight (BVLOS) operations, thanks to the presence of numerous sensors working at high frequencies (30–300 GHz, millimeter waves). These systems are currently being developed using different typologies of sensors (Yasin et al., 2020) (active or passive or combined) working in all operative conditions and overcoming the limits due to individual sensor usage.

These technologies are referred to as detect-and-avoid (DAA) systems that are designed to ensure safe flight of manned aircraft by detecting other aircraft in the airspace, identifying potential collision hazards and executing necessary maneuvers to avoid collisions with intruding aircraft (Yu et al., 2015; Besada et al., 2021). Despite the presence of transponders or radars in manned aircraft, mid-air collision avoidance is predominantly reliant on the human pilot's eyesight, as noted by the Federal Aviation Administration (FAA). It is well recognized that both human judgment and technological systems play crucial roles in the DAA processes for manned aircraft. In the context of UASs, DAA encompasses methods for detection and avoidance of collision with other aircraft or obstacles (Abro et al., 2022). This can be achieved either through onboard systems or with the involvement of a ground station (Ariante et al., 2022).

Typically, DAA sensor data include bearing angles (azimuth and elevation), range and relative velocity. A single sensor type often excels in measuring one specific data type. The complete relative trajectory is then calculated based on multiple sensors or additional data measurements. Commonly used DAA technologies for sensing include visual cameras, Global Navigation Satellite System (GNSS) sensors, Inertial Navigation System (INS) systems, thermal cameras, Light Detection and Ranging (LiDAR), radar and sonar (Moore, 2019; Ariante et al., 2019, 2020; Zhang and Hsu, 2018; Papa et al., 2018; De Haag et al., 2016; Alsalam et al., 2017).

However, UAVs predominantly operate in complex indoor, urban and confined environments where GNSS signals may be unavailable, and 5G connectivity might be insufficient for reliable data transmission. Consequently, the integration of on-board sensors, such as radar and other sensing technologies, becomes essential to ensure safe navigation (Lies et al., 2020; Siewert et al., 2020). The effectiveness of sensing and path-planning algorithms is strongly influenced by the computational capabilities of the UAV, the precision of its sensors and the degree of knowledge about the surrounding environment. Moreover, machine learning (ML) and neural network techniques have demonstrated significant advancements in various emerging fields, including robotic prediction and UAV operations (Back et al., 2020; Pedro et al., 2021; Lai and Huang, 2020; Fraga-Lamas et al., 2019).

The Texas Instruments (TI) IWR1642 is a millimeter-wave on-chip Frequency Modulated Continuous Wave (FMCW) Doppler radar, commonly used in the automotive industry. The novelty of this work lies in testing and characterizing the sensor for the development of a robust obstacle detection system tailored for UAM applications. Specifically, after the sensor's characterization, a dedicated structure was designed and developed to house the radar, as well as to accommodate additional sensors for future enhancements.

The primary objective of this study is to characterize the sensor's behavior under dynamic conditions. A preliminary investigation, based on static tests, was previously conducted by the authors to verify the sensor's maximum range, accuracy and precision (Menichino et al., 2023a, b). This work examines the dynamic behavior of the sensor in various scenarios (indoor and outdoor) and different chirp configurations, culminating in flight tests with the sensor mounted on board a UAV. The paper is organized as follows. Section 2 presents the main parameters of the FMCW radar and basic concepts of radar operation, and Section 3 provides a detailed description of the hardware components used. Section 4 presents the simulations and results obtained from ground-based campaigns. An in-depth error analysis is carried out in Section 5, followed by an illustration of the flight tests in Section 6. Concluding remarks and future developments are discussed in Section 7.

In millimeter-wave FMCW radar sensors, the transmitted signal is a linear chirp, i.e. a sinusoidal tone with linear frequency modulation, defined by the start frequency $fc$⁠, the bandwidth (B), the frequency slope S and the duration $Tc$ (Figure 1(a)). The time-domain expression of a linear chirp (also known as quadratic-phase signal) is:

for $t∈$ [0, Ramp End Time]; $ϕ0$ is the initial phase (i.e. at t = 0).

A single chirp can provide information about the object's range R by mixing (multiplying) the transmitted signal and the received echo (a chirp delayed by $τ=2R/c$⁠, where $c=3·108$ m/s is the speed of light in free space and, with good approximation, in air) and obtaining a “beat signal,” i.e. a sinusoidal tone with constant frequency Sτ (beat, or intermediate, frequency – IF) and initial phase equal to $2πfcτ$ (Figure 1(b)), or $4πR/λ$⁠, since $fc/c=1/λ$⁠, where λ is the wavelength. If multiple targets are present at different ranges, multiple chirps will be received, each delayed by a different amount of time proportional to their range to the sensors. The total received signal will contain different IF signals whose frequencies are measured via Fourier transform processing (the “range Fast Fourier Transform (FFT)”), deriving the targets’ ranges.

The configurable chirp parameters are.

1. Initial frequency (⁠$fc$⁠), which indicates the initial instantaneous frequency of the chirp. For IWR1642, it is equal to 77 GHz.

2. Slope (S) of the linear instantaneous frequency variation. For IWR1642, the maximum slope value is 100 MHz/μs. For a 40-μs chirp, the maximum chirp bandwidth is 4 GHz (from 77 to 81 GHz).

3. Idle time, i.e. the time between the end of one chirp and the start of the next one in a frame (approximately 7 μs for IWR1642).

4. TX start time, indicating the beginning of signal transmission.

5. Analog-to-digital conversion (ADC) valid start time, i.e. the time at which sampling begins.

6. Ramp end time, i.e. the time at which the linear frequency modulation stops. It is given by the sum of ADC valid start time, ADC sampling time and extra time after the end of the sampling.

7. Number of samples $Ns$ obtained by the ADC in the sampling interval.

8. Sampling frequency (⁠$fADC$⁠) of the IF signal, given by the ratio between the number of samples and the sampling time.

9. Chirp cycle time (⁠$Tc$⁠), i.e. the total duration of a chirp, given by the sum of the ramp end time and the idle time.

Detecting the velocity of a target requires the transmission of a sequence of N equally spaced chirps (a “frame”), with $N≥2$⁠. If N = 2 (single moving target at radial speed $v$⁠), the phase difference Δϕ between the two transmitted chirps will correspond to a range variation due to the motion of the object of $vTc$⁠, i.e. $Δϕ=4πvTc/λ$⁠, and the target radial speed is given by:

If multiple moving objects with different speeds and the same range are present, Δϕ will contain phase contributions from each target, and N equally spaced chirps (a frame) are transmitted. The speeds are derived by FFT of the N echoes (the “Doppler FFT”).

Range resolution (⁠$dres$⁠) refers to the smallest distance between two objects (with the same bearing and assumed equally reflective) at which the radar can distinguish between them. As is well known, $dres$ is related to the chirp bandwidth (⁠$B=STc$⁠) by:

where c is the speed of light. For 4-GHz chirp bandwidth, the range resolution is 3.75 cm.

Velocity resolution (⁠$vres$⁠) refers to the minimum radial velocity difference between two objects that allows them to be distinguished as separate targets by the radar system. This parameter is crucial for accurate tracking of moving objects. To avoid ambiguity in velocity estimation, condition <2π must be satisfied. From (2), this implies that the maximum unambiguous velocity is $|Δϕ|<2π$ when there is a single moving target. When transmitting a frame of N chirps, with duration $NTc$⁠, the velocity resolution is given by (Alabaster, 2012):

Angular resolution (⁠$ϑres$⁠) defines the minimum angular separation between two objects that allows them to be distinguished as separate targets and is dependent on the number of receiving antennas (⁠$NRX$⁠). Assuming, for example, two receiving antennas and a planar-wavefront observation geometry, the differential distance ΔR from the target to the RX antennas is given by $ℓsin⁡ϑ$⁠, where $ℓ$ is the spacing between the antennas and ϑ is the angle of arrival (AoA). The corresponding measured phase difference is $Δφ=2πΔR/λ=2πℓsin⁡ϑ/λ$⁠, from which the AoA is derived:

The estimation accuracy of the AoA ϑ depends on ϑ and is better when ϑ ≪ 1, so that $sin⁡ϑ≅ϑ$ and Δφ depends linearly on ϑ. To enhance $ϑres$ and improve the accuracy of angle measurements, Multiple Input Multiple Output (MIMO) radars are employed. These systems require at least two transmitting antennas (⁠$NTX≥2$⁠). By leveraging this configuration, a virtual array with $NRXxNTX$ elements is synthesized, effectively increasing the number of independent channels and improving the spatial resolution of the radar system. The angular resolution (in degrees) achieved by this approach is given by:

The effective detection of far-off objects is constrained by the signal-to-noise ratio (SNR) and the IF bandwidth supported by the radar system. The maximum range (⁠$dmax$⁠) of the radar system is directly related to the IF bandwidth as follows:

where $IFmax$ is the maximum IF bandwidth supported.

The setup (Figures 2 and 3) is composed by the following components.

1. AC-DC power supply;

2. IWR1642-BOOST board and

3. PC with TI mmWave Studio software.

Data processing involves different levels, with each level managing a radar subsystem (RADARSS), as illustrated in Figure 4, where.

1. Front-end processing. This stage entails the configuration of the sensor by the Demo Visualizer and the radar interface, interacting with the external environment through analog and digital front-end components.

2. Low-level processing. Here, the samples received from the ADC are processed by the digital signal processor (DSP). Range and Doppler FFT are performed.

3. High-level processing. Leveraging the microcontroller (MCU) ARM RF4, a cloud point is generated, representing the spatial coordinates of the targets. These coordinates are derived from the information obtained in the low-level processing stage and enable localization. Additionally, the velocity information for each target is obtained.

4. Post-processing. The processed data are further analyzed and displayed in the MATLAB^® environment.

The TI IWR1642 device (Figure 5) is an integrated single-chip mm-wave sensor based on FMCW radar technology, operating in the 76- to 81-GHz band with up to 4 GHz continuous chirp (Texas Instruments datasheet, 2024).

IWR1642 includes a monolithic implementation of an MIMO (2TX, 4RX) system and integrates a DSP subsystem (C674x) for radar signal processing. The device includes an ARM R4F-based processor subsystem, which is responsible for front-end configuration, control and calibration. Simple programming model changes can enable a wide variety of sensor implementations, with the possibility of dynamic reconfiguration for implementing a multimode sensor. Table 1 shows the performance values of IWR1642.

The TI evaluation board IWR1642-BOOST (Figure 6) is aimed to be used for the IWR1642 sensor performance assessment, enabling direct connectivity to the MCU LaunchPad™ Development Kit. The BoosterPack™ allows developing software for on-chip C67x DSP core and low-power ARM® R4F controllers, including onboard emulation for programming and debugging, as well as onboard buttons and light-emitting diodes for quick integration of a simple user interface. This kit is supported by mmWave tools and software, including mmWave Studio and the mmWave software development kit (MMWAVE-SDK) (Texas Instruments datasheet, 2024). The main features are.

1. Onboard antenna to enable field testing;

2. XDS110-based JTAG with serial port interface for flash programming;

3. Universal Asynchronous Receiver-Transmitter (UART)-to-Universal Serial Bus (USB) interface for control, configuration and data visualization;

4. TI LaunchPad interface to seamlessly connect to TI MCUs and

5. Controller area network (CAN) connector, which enables direct interface to target units.

The signal processing chain, visualized in Figure 7, consists of.

1. Radio frequency (RF)/Analog Subsystem (BSS): This subsystem incorporates all RF and analog components of the radar system. It includes transmitting antennas with corresponding power amplifiers (PA), a synthesizer, receiving antennas with low-noise amplifiers (LNA), mixers and ADC. The Baseband Signal Subsystem (BSS) interfaces the radar with the external environment, handling the transmission and reception of signals, as well as initial signal processing.

2. Digital Signal Subsystem (DSS): The DSS executes algorithms to estimate the range and velocity of detected targets. This is achieved through the application of range FFT and Doppler FFT processing, followed by an angle-FFT to determine the AoA of the signals.

3. Master Subsystem (MSS): The MSS oversees the operation of the radar, incorporating a Cortex-R4F microprocessor and various peripherals, such as UART interfaces. These peripherals facilitate the transmission of processed data to an external personal computer for visualization.

4. Digital interface and built-in self-test (BIST): Positioned between the BSS and the DSS and/or MSS subsystems, this block serves as the bridge connecting the analog and/or RF components with the digital signal processing sections. It ensures seamless communication between the signal transmission and/or reception components and the data processing and display functionalities.

In order to study and evaluate the operational characteristics of the sensor used in the present work, the tests performed, the methodologies applied, and the results obtained will be illustrated in the following paragraphs.

The data sessions were conducted in an outdoor scenario, utilizing a moving car as a target. This experimental setup was designed to simplify the initial testing phase, enabling a high frequency of trials with minimal resource expenditure. The choice of a vehicle as the target was strategic, as it provided a substantial radar cross-section (RCS), typically around 10 m² for a standard car (Kamel et al., 2017). This larger RCS significantly improved the probability of detection, facilitating the evaluation of system performance under realistic conditions. All sessions were held in the CIRA (Italian Aerospace Research Centre) facility, making use of outdoor spaces free from nearby metallic objects. The tests were classified into two types.

1. Along-track: The radar was positioned in front of the car, which moved towards and away from the sensor in a radial direction (Figure 8(a));

2. Cross-track: The car moved perpendicular to the radar's position (Figure 8(b)).

In both cases, a Geographical Positioning System (GPS) sensor was employed as a reference measurement onboard to ascertain the velocity and position of the car during the tests, facilitating a comparison with radar data.

To conduct these tests, four different chirp configurations were evaluated. Table 2 provides details on the maximum range and detectable velocities, as well as range and velocity resolutions and the number of antennas used for transmission and reception. Table 2 lists the chirp parameters corresponding to the four used configurations.

The frame duration was set at 100 milliseconds, ensuring that sampling was equal to the maximum allowed sampling frequency of 6.250 Mksps (see Table 3).

These tests aim to evaluate radar precision, velocity and range estimation by using a moving object along the radial direction toward and away from the sensor. The radar measures the radial component of the target velocity by exploiting a Doppler shift. Table 4 shows the tests performed and the corresponding chirp configurations used. Each test had a duration of about 30 seconds.

Due to the ground proximity, multiple tracks are observed corresponding to stationary objects and unwanted reflections from the ground, which were classified as clutter and removed. Figures 9 and 10 show the collected echoes and scatter plots (i.e. the position of the echoes in a XY-plane with origin at the antenna position) tests 1 and 7, respectively.

Figure 11 presents the velocities and distances of the target during test 1, obtained from the radar. Furthermore, Figure 12 depicts a comparison of velocities and distances derived from the GPS sensor for Test 7.

During Test 1, illustrated in Figure 11, when the target exceeded the maximum detectable velocity (8.4 m/s or 30.2 km/h, see Table 2), an instantaneous negative velocity value was observed, while the distance curve gradually increases. This is due to the target speed becoming greater than the maximum unambiguous speed, causing aliasing in the Doppler FFT and giving a folding of the measured velocity into a negative value.

Figure 13 shows the experimental setup. Two distinct configurations were selected based on the maximum and minimum values of the maximum radial velocity observed among the previous tests. This selection allows assessing the radar's capability to detect objects moving at various maximum velocities. Table 5 presents the conducted tests along with the corresponding chirp configurations utilized. Each chirp configuration underwent two tests: in Tests 1 and 3, the car moved from the left to the right side relative to the radar position, whereas in Tests 2 and 4, the movement was in the opposite direction.

Figure 14(a) shows the radar-target distance variation during the car motion, whereas Figure 14(b) depicts the scatter plots for Test 1, during experimental tests in cross-track modality. Figure 15 indicates radar-target distances during the Test 1 performed with chirp in Configuration 2.

Figure 16 shows the velocities and distances of the target, calculated from the radar, whereas Figure 17 shows the comparison of the distances between the radar and the on-board GPS sensor, during Test 1 and Test 2, respectively.

In post-processing, it was observed that the moving target was detected in all tests within an angular sector of 60-degree width (±30°), rather than the entire field of view of the sensor, which is 120° (±60°), due to the specular reflection of the metallic surface of the car, which, when viewed from an offset angle greater than 30°, backscattered extremely low energy toward the sensor. Moreover, within the time frame when the target was inside the radar's field of view, the distance values were coherent with the radial velocity values. Specifically, when the object was in the approach phase, the velocity values were negative (as described in the previous section). When the target was directly in front of the radar, the detected radial velocity was zero (Doppler shift equal to 0).

The data analysis and error evaluation were conducted in the MATLAB^® environment by comparing the tracking curves obtained from radar and GPS. Error analysis was specifically focused on along-track tests, as these provided a larger dataset for comparison between radar and GPS sensors. To compare the range curves, a fitting process was implemented. Specifically, two interpolating polynomial curves of appropriate degree were defined to provide a better approximation of the distance measurements obtained from the radar. To achieve a perfect match between radar and GPS data, it was necessary to synchronize the measurements before the interpolation process. After the synchronization phase, two polynomial curves were generated: the first curve was derived from radar measurements, and the second curve was obtained from GPS data, both at the same acquisition times. Figure 18, Figure 19 and Figure 20 depict the comparison between the polynomial curves for radar and GPS data for Test 1, Test 4 and Test 7, respectively, using a second-degree polynomial for Tests 1 and 4 and a fourth-degree polynomial for Test 7.

From the difference between the radar and GPS fitting curves, the absolute error and percentage relative error were calculated. The obtained results indicated that the radar can detect a target along a radial direction, with accuracy increasing as the distance between the radar and the target increases. These findings are consistent with the results obtained under static conditions (Menichino et al., 2023a, b). During the entire data campaign, the percentage relative error was generally less than 1%, with a maximum value of 3.7% observed in only one test (Test 1). The maximum absolute error estimated was 2 meters, while the mean absolute error was approximately 20 cm. The minimum absolute error recorded was 7 cm, which was estimated during test 3. The mean values of percentage error (Ε) and absolute error (|Ε|) for each test performed are shown in Figure 21 (see Figure 22 and Table 6).

In the comparison among velocities, a quantization error is present due to the difference between the analog sinusoidal signal and the quantized digital value, which is typical of the A/D conversion process. For this reason, the velocity curves show a “step” pattern when compared to the “smooth” curves from the GPS data. This discrepancy results in errors that display a “saw-tooth” trend. For representative purposes, Figure 23 shows the comparison between radar and GPS velocities during Test 5 in the along-track modality, highlighting the sampled radar velocity values.

Finally, after the sensor's characterization, to simulate autonomous operations in a UAM scenario, the radar was mounted onboard the UAV for flight tests. Figure 24 illustrates the structure designed and developed for this purpose, with the radar sensor in place. The flight tests were made possible through collaboration with ISARail S.P.A, a certified drone operator responsible for conducting the flight operations. The selected site for these tests is shown in Figure 25. All necessary procedural steps for the flight operations, including obtaining clearance for operations within the Grazzanise CTR (Southern Italy), were coordinated by the airfield manager. ISARail contributed to the process by conducting a detailed risk assessment and implementing mitigation measures to ensure the safety and effectiveness of the tests.

Figure 26 shows the system mounted onboard the UAV during the (a) calibration phase and during the (b) subsequent flight test. The calibration phase was crucial for ensuring that the radar system was properly aligned and configured to collect accurate data, while the flight test demonstrated the system's real-time performance in a dynamic environment. A single-board computer (SBC) Raspberry Pi 4 was used to receive data from the sensor to send them to the ground control station (GCS).

This dual-phase approach ensured the sensor's readiness for reliable obstacle detection and navigation during autonomous UAV operations.

During the flight tests, telemetry data were received in real-time to monitor and verify the functionality of the system. The data link used to receive telemetry from the onboard obstacle detection system consisted of a pair of transmitter and receiver units (Radiomodem SFMN) operating in the very high frequency/ultra high frequency band at 433 MHz. The transmitter, installed on the drone, was connected to a Raspberry Pi 4, while the receiver was linked to a laptop functioning as the GCS. This setup allowed for real-time data visualization and analysis during flight operations.

Based on the distance definitions reported in Figure 27, Figure 28 shows the points detected by the radar during the flight.

Figure 29 shows details of the minimum distance detected from the radar versus the distance of the hangar calculated from the GPS position of the trajectory.

It is possible to observe that the distances measured by the radar follow well the minimum (⁠$dmin$⁠) and maximum (⁠$dH1$⁠) distances from the hangar. There is a background noise with detection at about 5 meters due to the reflection of the ground and dependent on the drone's low altitude and flight pitch.

The result is that the radar can detect the hangar's wall at a distance starting from about 45m. In particular, the radar can correctly detect the minimum distance from the hangar's wall.

The following figures show the error between the distance measurement from the hangar detected by the radar and that calculated from the GPS coordinates of the trajectory.

In these figures, it is evident that the error is of the order of GPS measurement error. Therefore, the error is mainly attributable to the GPS measurement of the drone's position rather than to the radar, which in previous tests has shown to have accuracy in the order of cm (see Figure 30 and 31).

Figure 32 shows the detection of the radar at a point of the trajectory where the ground reflection is missing.

The tests were conducted using four different chirp configurations, with an onboard GPS sensor used as the reference measurement. The performed tests provided good results in terms of precision and accuracy of the sensor during dynamic motion, based on the four different chirp configurations used, set up by the mmWave Demo Visualizer software and exploiting the algorithms already implemented on the radar sensor. In the post-processing phase, the radar measurements were compared to GPS data in terms of position and velocity. The comparison of the tracks demonstrated a good match between the two sensors and confirmed their coherence with the target's trajectories. The mean absolute error evaluated was approximately 20 cm, and the percentage relative error was generally less than 1% on average throughout the entire data campaign.

After evaluating the sensor's behavior across different chirp configurations in several ground-based tests, flight tests were conducted to simulate operations within an UAM scenario. A dedicated structure was designed to house the radar, which was mounted on board the UAV. The flight tests were carried out at an airfield, using the hangar on-site as a simulated obstacle along the UAV's trajectory. The results demonstrated that the radar was able to detect the hangar's wall from a distance of approximately 45 meters. Notably, the radar consistently and accurately detected the minimum distance between the UAV and the hangar's wall, validating its performance in obstacle detection.

This paper evaluated the accuracy and precision of the TIIWR1642 radar sensor for UAM applications. The primary objective was to characterize the behavior of the sensor under dynamic conditions by performing two test campaigns: along-track and cross-track tests, relative to a fixed radar position. A moving car was chosen as the target due to its geometric shape and composition.

The application of the considered sensor appears to be suitable for obstacle detection systems in UAM operations.

Future work will include additional tests to improve radar performance using different chirp configurations. The primary objectives will be.

1. To develop configurations that increase the maximum detectable distance in both static and dynamic conditions;

2. To achieve maximum radial velocities on the order of 90 km/h by using the radar in Single-Input Multiple-Output mode with a single transmitting antenna and

3. To evaluate the detection of multiple targets (both fixed and mobile), with a focus on range and velocity resolution.