We have been working on low-speed unmanned delivery vehicles for over two years, mainly operating in three scenarios: parks, campuses, and factories. From the initial makeshift use of 2D radar and ultrasonic sensors, we have gradually transitioned to a full 3D perception system, encountering quite a few pitfalls along the way. Currently, the M360 from TanTu Zhixing is installed on the vehicle's roof, and we have previously tested the Livox MID-360 for comparison. Let's talk about our actual experience.

Working conditions for unmanned delivery vehicles

Let's start with the background. Our unmanned delivery vehicle is a box-type small car, about 1.2m long, 0.8m wide, and 0.9m high, with a load capacity of around 50kg and a speed of 0.5~1.5m/s. Its main purpose is to deliver packages, meals, and documents within the park.

Compared to AGVs and forklifts, unmanned delivery vehicles have several unique characteristics:

  • The route is not fixed. AGVs follow fixed channels, while delivery vehicles need to navigate between park roads, sidewalks, squares, and parking lots, with different routes every day.
  • There are many types of obstacles encountered. Pedestrians, bicycles, pets, shared bicycles on the ground, temporarily placed trash cans, construction barriers, etc., are all present.
  • There are a large number of unstructured scenes. There are no shelves or marking lines, and everything relies on perception and positioning.
  • Frequent switching between indoor and outdoor. From the lobby of the office building to the outdoor road, there is a significant change in lighting conditions.
  • They share the road with pedestrians. Delivery vehicles share the road with pedestrians on sidewalks or park roads, and the safety requirements are much higher than those in enclosed warehouses.

These characteristics determine that the requirements for radar in delivery vehicles are different from those discussed in the previous articles.

From 2D with ultrasonic to full 3D

The earliest batch of vehicles used a 2D single-line radar and ultrasonic sensor solution. The 2D radar was responsible for horizontal obstacle avoidance, while the ultrasonic sensor was responsible for close-range collision avoidance. Although it works, there are many issues.

The biggest problem with 2D radar is that it cannot see height information. There was a time when the vehicle was running on the park road, and the 2D radar showed that there was no obstacle ahead, but it collided with a branch sticking out from the roadside. The branch was just above the 2D scanning plane, and the ultrasonic sensor was installed below the front of the vehicle, which was not high enough. Since then, we have started considering 3D radar.

After switching to 3D, branches, suspended signs, and the rear of SUVs parked on the roadside that are higher than the vehicles can all be scanned. Ultrasonic sensors can be removed, and the overall sensor configuration becomes simpler.

Comparison of M360 vs. MID-360 on delivery vehicles

Both radars are installed in the center-forward position on the roof, about 0.9m above the roof, inclined downward by 10°.

Blind zone: 5cm vs 10cm

The ground clearance of the delivery vehicle's chassis is about 15cm. Common low obstacles on the road surface—curbs, speed bumps, fallen shared bicycle pedals, pets—generally range in height from 5 to 15cm.

A radar with a 10cm blind zone is almost ineffective against obstacles between 5~10cm. We tested a 200m park road, where the M360 avoided 8 low obstacles, and the MID-360 avoided 6. The two missed were a fallen shared bicycle pedal (about 6cm high) and a piece of broken brick at the edge of the curb (about 4cm).

The 6cm pedal may not seem like a big deal, but the delivery vehicle's undercarriage has only a 15cm ground clearance, which could potentially jam the undercarriage. The 4cm piece of broken brick is unlikely to cause damage, but the vibration from running over it could affect the stability of the goods inside the vehicle.

For delivery vehicles, a 5cm blind zone is not absolutely necessary, but it can indeed help avoid more troubles.

Vertical FOV: 70° vs 59°

This difference is more pronounced in the context of delivery vehicle scenarios.

The delivery vehicle is mounted on the roof, with a height of less than 1m and inclined downward by 10°. It needs to cover both the ground (detecting curbs, steps, low obstacles) and the overhead space (trees, signs, vehicle rear ends).

M360's 70° vertical FOV, installed at a 10° tilt at a height of 0.9m, theoretically covers an area from 0.3m above the ground to about 1.5m up. It can basically cover the three-dimensional space around the delivery vehicle.

MID-360's 59° coverage range is 11° narrower, reaching only about 1.2m up. Some taller obstacles—such as the rear of an SUV, branches—may just be on the edge of the scanning range, with insufficient point cloud density.

Actual experience: The M360 is more stable in detecting overhead obstacles, especially on roads in parks with many trees, where the performance gap is quite obvious.

Weight and Dimensions

The MID-360 is 143g lighter than the M360 at 265g.

Delivery vehicles are small, with the entire vehicle weighing about 80kg. At 408g, it doesn't significantly affect the center of gravity, but if pursuing extreme lightweight (such as in scenarios with weight restrictions), the lightweight nature of the MID-360 has an advantage.

In terms of size, the M360's 78×78×81mm is larger than the MID-360's 65×65×60mm. In cases where there is limited installation space on the roof, the MID-360 is easier to fit in.

Power consumption

Delivery vehicles usually use a 24V or 48V battery pack with a total capacity of 30~50Ah. The overall power consumption of the vehicle is approximately 100~200W (driving motor + computing platform + sensors).

M360's 4.5W vs. MID-360's 6.5W, a difference of 2W. Running for 12 hours a day is 24Wh. For a 50Ah 48V battery pack, this is less than 1%.

Power consumption is not the deciding factor in choosing the radar for the delivery vehicle. However, if there are many sensors on the whole vehicle (3D radar + camera + ultrasonic), every watt saved adds up.

Power supply range

Our vehicles are powered by a 24V battery pack directly supplying power to the sensor. The M360 is perfectly compatible with 12~32V, and the MID-360 can also work with 9~27V.

However, in low-temperature environments, battery voltage will drop. In winter, when it's zero degrees, a fully charged 24V lead-acid battery may only be around 25V, and it may drop to 23V when half charged. Although MID-360's 27V upper limit is fine at this voltage, the margin is indeed much less than M360's 32V.

We had a vehicle that reported an error due to low battery voltage when starting the MID-360 after being frozen overnight in winter. It worked normally after recharging. The M360 has never had this issue.

Dual echoes

Delivery vehicles spend a lot of time outdoors, and dual echoes are indeed useful in rainy weather.

We have a shaded path in our park, and on rainy days, the wet road surface and rainwater from the trees on both sides will drip onto the road. In single echo mode, there are more raindrop noises, especially on sections with dense trees on both sides, where the point cloud is full of noise.

After enabling dual echoes, raindrop noise is captured and filtered by the forward echo, while the backward echo retains actual obstacle information. The outlines of trees on both sides, road speed bumps, and parked vehicles can all maintain clear point clouds.

However, we usually slow down or stop operations during rainy days, and the dual-echo is mainly for safety redundancy.

Non-repetitive Scanning

Delivery vehicles move slowly and stop at traffic lights or when waiting for pedestrians to cross. Non-repetitive scanning accumulates point cloud density in a stationary state, greatly aiding in perception of intersection scenarios.

Intersections are one of the most complex scenes for delivery vehicles. It is necessary to pay attention to: vehicles approaching from the left and right, pedestrians crossing the road, curbs, and manholes on the road surface. Both the M360 and MID-360 support non-repetitive scanning. After stopping at the intersection for 2~3 seconds, the density of the surrounding environment's point cloud increases significantly, making the detection of small obstacles more reliable.

IP67

Delivery vehicles operate outdoors, enduring rain, dust, and mud. IP67 is a must-have. Both companies now offer IP67, making it a level playing field.

However, there's a detail: the radar on the delivery vehicle is exposed on the outside, without any bodywork to shield it. With direct sunlight and working heat, the ground temperature in summer may exceed 60°C. The operating temperature range of the M360 is -10°C to +60°C, while the MID-360 is -20°C to +55°C.

MID-360's high-temperature limit is 5°C lower. In extreme hot weather, M360 has a larger margin. We haven't encountered any radar shutdown due to high temperatures in summer, but this difference is worth noting if doing outdoor delivery in southern cities.

Our current solution

All 6 park delivery vehicles are equipped with M360 (including 2 M360-D dual-echo versions).

Sensor configuration:

  • One M360 on the front of the roof, height of 0.9m, inclined 10° downwards
  • One ultrasonic sensor at each corner of the vehicle (for close-range collision avoidance)
  • 1 wide-angle camera in front (pedestrian recognition + traffic light recognition)
  • Positioning solution: 3D LiDAR + RTK-GNSS + IMU

The main considerations for choosing the M360 instead of the MID-360 are: a 70° vertical FOV that covers more completely, a sufficient margin for starting voltage in low temperatures, and dual echo redundancy in rainy weather.

If there is an extreme requirement for volume and weight (such as smaller end-of-line delivery robots), the MID-360 is a reasonable choice.

Selection recommendations

When selecting 3D radar for unmanned delivery vehicles, several parameters are worth focusing on:

  • Vertical FOV: As large as possible, low installation height on the roof, covering the entire upper and lower space
  • Near-field blind zone: Below 5cm is better, as there are many low obstacles on the road surface.
  • Operating temperature: Pay attention to the high-temperature limit, as it can be very hot outdoors in the summer sun.
  • IP67: Essential for outdoor use
  • Non-repetitive Scanning: Accumulate point clouds while waiting at intersections to enhance perception reliability.
  • Weight and Dimensions: Small vehicles are sensitive to weight, so they should be as light as possible.

Both M360 and MID-360 can meet the basic needs of delivery vehicles. M360 has an advantage in FOV and power supply, while MID-360 has an advantage in size and lower temperature range. It depends on the size of your vehicle and the environment you operate in.

*The above is based on actual testing experience. The parameters of MID-360 are subject to the latest release on the Livox official website, and the parameters of M360 are subject to the official information of TanTu Zhixing.*