Foresight Autonomous Holdings — Update 27 September 2016

Foresight is a development stage company using the know-how of parent company Magna BSP, developed with the Israeli military, to develop algorithms for forward stereo camera systems in vehicles. Its aim is to achieve 100% object detection in all weather and light conditions – well ahead of company estimates of current industry norms of c 85%. If successful FRST should be able to build market share in both the ADAS and FA vehicle markets, the latter with the planned launch of sensor fusion software in 2020. The company is looking to accelerate its market entry with a focus on the large and fast-to-market Chinese auto industry where the authorities are looking to bring safety standards to EU levels by 2018. After planned warrant exercises, FRST will shortly have a 31.5% stake in Rail Vision Ltd, which is currently the only known company developing an ADAS-type solution for the rail industry. Its system has a 1.5km object detection range for night/day/all-weather use using far infrared and CCD cameras. In addition to sales of the systems, RV will generate revenues from data collected by sensors on the trains. RV’s technology has already undergone significant testing by Israel Railways and its proof of concept (POC) is due in Q317.

Our forecasts are based on milestone achievements (see Exhibit 1), and penetration of the new car market by forward cameras in both ADAS and increasingly the FA vehicle market rising from c 10% in 2016 to 73% in 2025 and FRST achieving 2025 market shares of 6% in China and 2.0% in the rest of the world. We forecast EBITDA to turn positive in 2020.

Monthly cash burn excluding equity investments at FRST is forecast at NIS880k ($230k) in H216 rising to NIS1.1m ($285k) in 2017. Cash at the end of H116 plus recent funds raised (c NIS4.7m) should fund FRST until Q118 including the planned $0.8m outlay to increase the company’s holding in RV to 31.5%. We believe that FRST has the potential to close most if not all of its NIS30.0m ($7.9m) funding gap until 2020 with the proposed $7-10m Nasdaq issue.

As FRST is a pre-revenue development stage company, we have valued it using a DCF analysis with a 15% WACC. Based on FRST hitting its key milestones and our market share and earnings targets (see Exhibits 1 and 9), we value FRST shares at NIS3.15. After the launch of the POC we see potential for this valuation to increase towards NIS4.52. This assumes the company is able to demonstrate sufficiently superior performance of its software versus that of its competitors, to justify revising 2025 target market shares from 6% to 8% in China and from 2.0% to 3.5% in the ROW.

FRST is a development stage business with potentially disruptive technology moving towards commercialisation. As such there are significant risks, uncertainties and potential. With the technology still unproven, successful launch of the POC is the critical near-term milestone. There is no guarantee that targeted object detection rates and lower accident rates will be achieved and FRST may struggle to gain targeted prices for its software in a highly competitive market. We estimate that the company will need to raise NIS30.0m ($7.9m) by Q118 but this may be satisfied by a planned Nasdaq equity raising later this year. Longer than expected time to convert pilot trials to volume sales or faster than expected price erosion could increase this shortfall. Software sales are typically denominated in US dollars, but the bulk of FRST’s costs are in Israeli shekels. Sales in China could be significantly higher if the country achieves its plan to meet Euro NCAP standards by 2018. EBITDA margins at RV could be higher than we model if it retains market leader position and generates significant big data revenues. We see potential for sector consolidation.

Foresight was created in 2011 to develop software to adapt parent company Magna BSP’s 3D multi-camera based intruder detection system for ADAS use. Magna BSP’s technology is deployed around the world in sensitive facilities such as airports, nuclear reactors and prisons. Its success is based on a claimed 99.9% detection rate in every terrain and climate using image and infrared (IR) camera surveillance. Magna BSP invested $2m into Foresight’s R&D and has transferred the rights to Foresight to use its video motion detection and object tracking know-how in the auto industry without charge. After reversing into a listed shell, FRST listed on the Tel Aviv Stock Exchange on 5 January 2016 and plans to launch a proof of concept of its first ADAS software with a targeted 100% object detection probability in daylight in Q317. In September, Foresight announced its intention to exercise warrants to increase its stake to a minimum of 31.5% in Rail Vision, a company that aims to use ADAS technology to prevent rail accidents as well as to create big-data revenues from on-train sensors. RV plans to start selling its first systems in Q317.

Foresight is looking to bring about a step change in object detection rates in vehicle forward vision systems using innovative algorithms to achieve object detection rates of 100% in all-weather/light conditions. The timing is good. Recent tests by German auto association ADAC showed surprisingly poor results by the AEB systems of six major marques (including the BMW 3 series, Volvo V60 and Mercedes C Class) in avoiding collisions, particularly with smaller objects such as cyclists and children. With regulators and OEMs currently promoting ADAS with a view to safety and product differentiation, before all else detection matters now. Going forward, one of the most pressing problems of OEMs developing FA vehicles is the need to build consumer confidence in self-driving cars in the face of recent high-profile accidents. Given this, we expect products offering significant safety advances to benefit from both market share gains and participation in the fast-growing market for vehicle automation.

Strategic objectives: Automotive (ADAS & FA vehicles), software-only product, global scope

Rail Vision: a complete solution, software and hardware; initially European/US focused

In Exhibit 1 and 2 we summarise the key industry and company milestones to track out to 2022.

Foresight aims to achieve a significant advance in the reliability of forward cameras with its first product launch in Q317. Using software to process stereo camera images, it aims for functional object detection of 100% (classing objects by categories, eg “pedestrians”, “cyclists”, “other small objects”), compared with its estimate of current industry levels of c 85%. It also aims for zero false positives to avoid unnecessary braking and swerving. Recent tests by the German auto association ADAC appears to show significant potential for improvement in AEB object detection systems in the market. Rating performance on a scale of 0 to 100% (100% being an untouched dummy 0% a run-over dummy), all but one of the six vehicles tested (the Audi A4) scored 25% or less in encounters with a cyclist at 40 km/h, and the same vehicles scored 54% or less testing for a partly obscured child at 50 km/hr. Subaru’s stereo system was best in class at night and with adult pedestrians.

We summarise FRST’s advantages/key advances as follows:

Most autonomous and semi-autonomous driving systems employ at least one image camera (a charge-coupled device [CCD]) and another sensor at the front of vehicles. Cameras are low-cost and enable the identification of objects. Radar is very good at calculating distance and speed of objects but cannot be used to identify objects. Lidar is very good at determining the position and shape of an object, but cannot be used to determine colour, signs and images and is very high cost. Within the camera space, mono cameras use a single camera to analyse the content of images, while stereo cameras operate like human eyes, using the parallax shift between images to accurately measure distance and size.

As can be seen in Exhibit 5, leading automakers BMW, Jaguar, Mercedes-Benz and Land Rover all use stereo camera technology in their top specification vehicles. New stereo housings in the latest Model X suggest that Tesla is also looking to switch to forward stereo cameras.

ADAS is one of the fastest growing markets in the auto industry. ABI Research forecasts total spending in the ADAS market to grow at a 29% CAGR over the next decade, with sales of vulnerable user (cyclists/pedestrians) detection systems, which are principally reliant on forward camera systems, forecast to achieve a 49% CAGR during this period. From 2020, the growth of FA vehicle production should further boost forward camera demand. Key market drivers, therefore, are:

Europe and in particular Germany lead the way in ADAS. As early as 2014, 20% of new German cars were fitted with AEB and 21% with lane keeping assist (LKA) functions (source: Bosch). In the US, in 2016 AEB is available in 51% of all vehicle models, but standard in only 6% of new cars (vs 7% in the UK). Volvo and Tesla were the only OEMs to offer it across their entire ranges (source: Jato Dynamics Ltd).

Research in China estimates global automotive forward camera shipments at 9.1m units in 2015. This implies global forward camera penetration of c 5-7% of new vehicles based on an average 1.5-2.0 front cameras per car. Based on regional estimates, we forecast global penetration of forward cameras in new cars at c 10% in 2016, rising to 36% in 2020 and 73% in 2025.

China is the largest global vehicle producer, with sales of 24.6m units in 2015, representing 27% of global output. Demand is being driven by low current levels of car ownership, increasing urbanisation and rapid growth in the proportion of middle-income earners. Positively for FRST, there are currently no local Tier Ones selling ADAS products in China.

Among developing countries, China has above-average standards in passenger car safety regulation. Seat belt use is mandatory and assisted braking systems, mandatory crash tests and airbags are all compulsory in new cars, well ahead of the requirements in India, Brazil or Mexico. With an eye on Western export markets, the Chinese authorities recently announced ambitious plans to catch up with the world-leading Euro NCAP standards by 2018. This would imply mandatory AEB for detection of pedestrians and cyclists for top-rated cars by 2018. Realistically, in view of the challenge this represents and reluctance among Chinese OEMs to link ADAS to brakes at present, probably owing to high-profile AEB recalls, we assume that such measures will be introduced in 2020-25 and mandatory AEB in all cars will take place in 2027-30 (vs 2022 in the US).

After regulation, we see affordability as the key factor governing ADAS take-up in the Chinese market. The country has a very high output of budget class cars; in 2015, 29% were priced at $12k or less. Nevertheless, many car owners are young professionals, for whom safety is a significant consideration, who have already bought entry-level cars and are looking to upgrade to larger vehicles. In 2015, 31% of cars sold were in the $22,500 or higher price category and McKinsey forecasts this to grow to 33% by 2020. Helped by falling hardware costs and rising economies of scale, we see this as the key addressable market for well-priced ADAS stereo camera systems, on a five- to 10-year view, taking into account the $26k starting price of stereo-camera-as-standard Subaru cars in the US. Assuming 30% penetration of this market and full penetration of the luxury market (forecast at 10% of sales in 2020), by 2020 we forecast forward camera based ADAS penetration to reach 20% of new car sales in China, rising to 60% in 2025.

Product pricing and positioning

FRST plans initially to price its basic and premium software (including IR) at $45 and $90, respectively. The company estimates that by 2020 when it should be selling in volume in China the cost of a pair of CCD stereo cameras should have fallen to c $150 (vs c $230 for the CCD/IR stereo cameras for the premium software). Adding platform costs of c $50 and FRST’s basic software (by then forecast at c $30) should enable OEMs makers to install high-quality stereo ADAS for c $230 per car, representing c 1% additional cost in the addressable market. It is worth noting that Mobileye’s third-generation EyeQ3 mono system on a chip currently sells at c $50, so with a c $100 mono camera costs c $150, and so FRST is not looking to undercut Mobileye, but win on quality.

We accordingly expect FRST to position its products as:

Foresight is looking to work primarily with Tier One and Tier Two auto suppliers to bring its product to market. This has a number of advantages; by placing its software on an existing Tier One platform the company is not encumbered with concerns about attaining ASIL (automotive safety integrity level) standards or hacker/security issues. Additionally, the time to market is considerably lower than working directly with OEMs as new software can be integrated into a Tier One’s platform and introduced as an upgrade to an existing vehicle in relatively short time.

With the intention to enable as wide a possible addressable market of Tier Ones, FRST is working ahead of the launch of its POC to make its algorithms available on at least four commonly used chipsets compatible with the platforms of a wide array of Tier One suppliers. It is worth contrasting this approach with that of market leader Mobileye, which has an estimated 60-70% market share in mono camera ADAS, and sells its software only on its own proprietary chip. We believe that Mobileye’s recent agreement with Tier One Delphi to create a complete “off the shelf” ADAS platform for OEMs could shrink the addressable market for independent ADAS software if a high proportion of auto manufacturers adopt its system.

The Tier One and Tier Two market is global but, reflecting its fragmented nature, highly competitive. According to IHS Automotive, in 2015 65% of light vehicle production was supplied by global platforms, but major OEMs such as Ford typically deal with over 1,000 authorised suppliers.

As can be seen in Exhibits 3 and 7, a number of Tier Ones already have significant engagement with stereo, so there is considerable potential competition in the market from external and in-house software developers. Nevertheless, we still see them as part, if not a key part, of FRST’s addressable market. As mentioned earlier, the key challenge for FRST will be to demonstrate that its solutions represent a substantial step forward in safety terms, which it is working to do via extensive driving hours. If it does this, then Tier Ones already supplying OEMs with stereo rigs in their models are likely to be a receptive audience in their quest to retain these OEMs as customers.

Tier One Autoliv and OEM Subaru both claim a similar c 100m stereo camera range to FRST, so the company does face direct competition in longer-range stereo. While it is not possible to exclude the possibility that other Tier Ones are aiming for similar improvements in object detection rates to those targeted by FRST, we believe FRST has the advantage of access to Magna BSP’s 12 years’ know-how in stereo camera object detection for better interpretation of the infrared and image (CCD) camera output).

Rail Vision is a development stage company working on a collision avoidance system using thermal camera imaging and military technology. The aim is to detect objects in the path of trains and apply autonomous braking systems, if necessary. In addition, it will incorporate sensors and cameras into the on-board system to collect and sell a range of data from the trains’ surroundings including weather metrics, network mapping, infrastructure conditions, etc, to generate recurring revenues.

The business was founded in 2014 by four scientists: Elen Katz, CEO, who has a robotics background with Homeland Security; Shahar Hania (VP, master’s physics/electro optics) and Dr Yuval Isbi (PhD physics/electro optics), both with backgrounds at defence contractor Elbit Systems and Noam Taich (COO) with extensive experience in complex stills and videos in defence. RV was incorporated in May 2016 and FRST acquired a 13.5% stake in August 2016 for NIS2.3m ($0.6m). In Q416 FRST plans to raise this to a minimum 31.5% by exercising warrants worth at least $0.8m, which will also result in it having warrants to increase the stake to 47% by February 2019 for a further $11.3m.

RV was founded to apply ADAS solutions to rail systems, and in particular the problems arising from the extreme high speeds of modern trains vs slow reaction times of humans. At 300-400km per hour, two seconds of thought can see a train move 200m along the track before brakes are even applied. According to the company, many attempts to solve the problem with video cameras have failed because they do not work at night time or in low visibility. RV employs thermal (far infrared) and CCD (image) cameras mounted on top of a driving unit to enable the system to “see” up to 1.5km ahead (in clear line of sight) and autonomously apply brakes while alerting the driver and central controllers of a dangerous object ahead. While the CCD camera provides rich data in the daytime, the infrared camera picks up the different thermal signatures of objects, including “cold” concrete, which has a heat factor 50 times that of rails. This enables the system to operate at night time and in poor light and a range of severe weather including fog, sand, dust and snow.

RV is currently approaching the leading European and North American railway operators with the aim of selling complete systems (hardware and integrated software) directly to them. Management believes that it has first-mover advantage in long-range detection based on its discussions with leading European rail companies. We understand that Siemens has worked on a similar idea but achieved only sub-100m camera ranges similar to that of Bombardier’s system for light rail (trams). RV’s system has successfully completed six trials with Israel Railways and this month started a pilot trail with the Italian national rail operator, FS Italiane, for safety and network mapping services. RV is also exploring an autonomous train pilot with Rio Tinto and an alliance with GE Israel for infrastructure inspection. The company plans to release the proof of concept for its collision avoidance system in Q317 and to achieve sales of several units shortly thereafter to either Israel Rail or FS Italiane. In its POC it intends to demonstrate the following KPIs:

To date camera and sensor technology has only been located at key points on the rail networks (eg level crossings) and under track monitoring equipment. Siemens is currently developing a system placing sensors along rail lines to sense objects in the track area and halt trains in the area, if needed. We can see considerable utility for the system in smaller EU rail networks, but believe that capital costs could act as a barrier to widespread application in larger and poorer countries.

RV expects the greatest initial demand to come from highly safety-oriented Europe and the US. Looking at the case for the EU, based on the expected initial $50k unit cost and assuming $5k installation, we estimate a cost of $6.6bn to install the system across the EU powered rail unit (PRU) fleet. According to the European Railway Agency, significant rail accidents cost €2.5bn ($2.8bn) in 2011. In over 85% of cases, the accidents were caused by collisions with objects, not other trains. On this basis, a 25-50% reduction in these collisions would give rise to a 9-18% ROI. The planned $50k module price can also be equated to a fairly low 1.5-2.5% insurance policy against the typical $2-3m overall cost of locomotives and potentially high cargo values.

We believe that a stronger case for such systems can be made in large-scale networks in Russia, China, India and Brazil. Unguarded crossing points and signalling problems pose significant dangers of accidents and derailments. On this basis, we expect initial orders from Europe and the US to be quickly followed by demand from Russia and Asia.

We see sales of data as being a key part of the business model. With many rail networks crossing remote areas, sensors on trains have applications including weather monitoring, track assessment and mapping of infrastructure. RV intends to actively pursue revenue streams from this area.

We have modelled the market for Rail Vision’s modules based on an estimate of the proportion of the existing global PRU fleet that will adopt ADAS systems. We assume that RV will launch its product in Q317, with the immediate sale of five units probably to Israel Rail or FS Italiane. We assume an initial 50% addressable market as RV will focus initially on Europe/North America, rising to 100% in 2020 as the sales effort expands. See Exhibit 8 for other key revenue assumptions.

We have conservatively forecast EBITDA margins at Rail Vision of 15.0% from 2020, representing a 75% weighting of tech hardware sector EBITDA margins of 10.6% over last four years and a 25% weighting of software and semiconductor margins of 27.7% over the same term. We see the potential for RV to achieve higher margins if it maintains market leadership and is able to enhance hardware sales with higher big data revenues. We have also forecast big data revenues per ADAS-installed PRU at $500 per annum, which is less than half of management expectations based on our concern that some rail companies may seek to retain a share of such revenues, or forbid the use of sensors on the trains. Based on the above we derive a valuation of the company of NIS129.2m ($34.2m), equivalent to NIS0.55 per diluted FRST share.

As of mid-2016 we understand that the company had cash of c $700k (with no debt). According to FRST management, cash burn since incorporation in May is currently averaging c $100k/month. First sales revenues are expected in Q317 but we see a potential funding need to EBITDA break-even in 2020 of NIS31.8m ($8.4m). A further $0.8m will come from FRST as it has announced plans to exercise at least this value of warrants to bring its stake to at least 31.5% in Q416. More funding may also come from other equity injections, with FRST having warrants to potentially increase its stake further to 47% by February 2019 for an additional $11.3m.

The management team at FRST is headed by Haim Siboni, the founder and CEO of both Foresight and parent company Magna BSP. Since the foundation of 2001, Mr Siboni has led Magna BSP to successfully develop stereo camera technology to achieve 99.5% object detection accuracy and built a leading position in global facilities and border security. COO and Co-CEO Ariel Dor has more than 10 years’ experience in technology management, including telematics firm Galooli Ltd, and defence technology firm Elbit Systems Ltd. CTO Levy Zruya joined Magna BSP (where he continues as CTO) at its inception and is an engineer with 30 years’ experience in high-tech industries, including the Israel Defense Forces and Israel Aircraft Industries. He is an expert in real-time software development, image processing and sensors. CFO Eli Yoresh has over 15 years’ financial experience and an automotive background, having formerly been CEO of Tomcar Global Holdings, an international manufacturer of off-road vehicles.

Our forecasts and valuation are affected by the following sensitivities:

FRST was incorporated at the start of 2016 with the reverse takeover of a listed Israeli company. The H116 results were therefore dominated by merger related expenses. The group reported an EBITDA loss of NIS43.0m, which after exceptionals of NIS33.2m and share-based expenses of NIS6.2m produced a normalised EBITDA loss of NIS3.6m.

Our forecasts for FRST are dependent on the company reaching a number of key milestones as shown in Exhibit 1, and discussed in Exhibit 2 above. The most important event for our model is the planned launch of the proof of concept for the Basic and Premium versions of the ADAS software in Q317. We forecast FRST to become profitable at the EBITDA level and to start generating free cash flow in 2020, later than management, which is looking for positive PBT by end 2019.

Our key revenue and earnings assumptions are shown in Exhibit 9. We assume:

Helped by share issues raising a net NIS18.6m ($4.9m) in H116, FRST finished the six months to 30 June with NIS19.4m ($5.0m) cash and no financial debt. With the recent additional NIS4.7m ($1.25m) capital raising earlier this month at NIS1.75 per share, our model indicates that the company has sufficient funding to take it through to Q118. This is based on an estimated cash burn for H216 and 2017 of NIS4.8m ($1.3m) and NIS12.9m ($3.4m), respectively, plus the NIS5.3m ($1.4m) outlays on increasing its stake to 31.5% of Rail Vision. We estimate a further funding requirement in 2018 and 2019 of c NIS30.0m ($7.9m) before the company becomes self-funding from 2020. FRST has announced plans for a Nasdaq listing and capital raising of $7-10m later this year. Furthermore, the company currently has 12.6m warrants outstanding with an exercise price of NIS3 that mature in H217, which if exercised would bring in a further NIS37.8m ($10.0m).

FRST had cash reserves at mid-2016 of NIS19.4.6m ($5.0m) and NIS3.0m liabilities made up of working capital and a government grant.

As FRST is a development stage company with no readily comparable listed peers, we have employed a DCF model for our valuation (see Exhibit 10). Our key assumptions are:

This gives rise to a valuation of NIS3.15 per share, of which Rail Vision comprises NIS0.55 per share. Our optimum case of a 2025 8% market share in China and 3.5% ROW gives rise to a valuation of NIS4.52 per share. The sensitivity of the valuation to 2025 market share assumptions is shown in Exhibit 11 and its sensitivity to WACC and terminal growth assumptions in Exhibit 12.