Join us to participate in the upcoming 2019 City:One Challenge. 

User-centered travel behavior intervention technology to empower older adults to walk by providing truly personalized recommendations

Adaptive Artificial Intelligence as the foundation for personalized interventions to make walking accessible and attractive for older adults

Photo of Jose Mantilla
2 0

Written by

Why are we doing this?

Travel behavior change interventions have had no demonstrable significant and sustained reductions in car use (and associated increase in active travel and physical activity). This is primarily due to their intrinsic inability to provide truly personalized and realistic recommendations for change.

To date, most individual-level behavior change interventions to promote active travel and physical activity have yielded mostly disappointing results for a number of reasons: (1) require people to set aside time for exercise even though lack of time is consistently identified as a barrier to physical activity across all sociodemographic groups and it is well-known that people are much more likely to be physically active if ‘exercise’ is part of the normal routine, as it requires a lower commitment threshold); (2) largely based on generic and limited information (e.g., walk 10,000 steps per day) and thus unable to provide personalized and actionable recommendations for change; (3) ignore health conditions (e.g., diabetes), physical limitations (e.g., knee osteoarthritis), general capabilities (e.g., walking speed and endurance) and individual preferences; (4) ignore daily routine and everyday travel patterns; and (5) based on unreliable systems (e.g., wearable devices [such as Fitbit] and smartphone applications) that have been developed/calibrated for healthy populations (i.e., existing devices capture only around 50-60% of steps for severely obese individuals and around 65-75% of steps for older people).

After decades of efforts, collective scientific knowledge amounts to promoting broad common-sense lifestyle objectives lacking true insights into the practical means to deliver change. Promoting active travel and physical activity remain as key challenges facing transportation and health researchers and practitioners.

Why older adults?

The Centers for Disease Control and Prevention have recognized that physical activity in adults over 65 years old can help maintain healthy and longer lives, and can prevent many health problems that come with age such as heart disease, diabetes, and arthritis. Physical inactivity is a risk factor that contributes to the growing and significant burden of chronic disease around the world. Even though it is widely acknowledged that physical activity improves health and is a critical prevention tool to reduce the risk of chronic diseases and cognitive deterioration, the proportion and number of older adults in the United States (and many other countries around the world) doing insufficient, very little or no exercise has continued to increase.

Our solution

Our team will deploy a user-centered smartphone-based travel behavior change intervention program that will empower older adults to walk by providing personalized active travel and physical activity recommendations that are commensurate with their individual capabilities and embedded into their daily travel routines. The primary focus of our program is to transfer trips: (1) from motor vehicles to any other travel mode or combination of modes; and (2) from transit to walking. 

The platform includes two primary components: (1) activity detection system; and (2) travel behavior change platform. The integration of these two components will enable our team to provide older adults with fully personalized recommendations for change that are commensurate with their individual capabilities and embedded into their daily routines. The importance of providing realistic recommendations (in contrast to the vague and generic nature of information provided by all exiting travel planning tools and fitness devices) is that it will truly empower people to shift trips (or portions of them) from motor vehicles to sustainable modes.

Our team has already developed, in collaboration with a leading data science company, an adaptive Artificial Intelligence world-class activity detection system that uses information collected from smartphone sensors (e.g., accelerometer, magnetometer, gyroscope and GPS) to determine when/where/how/why people travel. More specifically, our system determines (to a high level of precision): (a) origins and destinations of all trips (e.g., home and school, home and work, or work and shopping); (b) routes traveled; (c) real door-to-door travel time for all trips; and (d) travel mode(s) used (i.e., whether people walked, rode a bicycle, traveled on bus, tram or train, or drove, or used a combination of two or more modes). This fine-grain and person-level information enables the identification of true/revealed travel patterns for each individual – taking the guesswork out of travel behavior change efforts. Only by knowing (to a high level of precision) what people do today, we can understand (and begin to influence) what people may be able and willing to do tomorrow.

In collaboration with the Motion Analysis Laboratory of the Harvard University Medical School, we have developed algorithms (through simulations and inference) that determine personalized/desirable/safe physical activity thresholds (i.e., maximum number of daily steps, minutes and kilometers per day and per individual event) based on people’s age, gender, weight, height and health condition. This is the result of simulations for thousands of patients over many years.

By combining the physical activity thresholds (what individuals can actually achieve in terms of walking) with information from the activity detection system (what individuals actually do on a routine basis), we identify those trips (for each individual person) that are realistic candidates for ‘change’ (based on trip characteristics, individual capability and time constraints). These trips are those for which it is feasible to: (a) switch from car (private, taxi, Uber/Lyft or shared mobility) to transit, bike or walking; or (b) switch from transit to walking. Only trips for which a shift results in time gain or minimal additional travel time are considered. For example, long trips with no feasible and efficient transit options are ‘ignored’, whereas short trips that are eminently walkable (or for which viable time-competitive alternatives are available) are ‘considered’. Importantly, since our system identifies the true origins and destinations of trips, as well as the subcomponents of trips, it also identifies opportunities for physical activity by switching portions of longer trips (e.g., walk the last segment of a trip to work or alight the tram a stop earlier and walk the ‘rest of the way’).

Our team has also developed a user-centered behavior change platform that includes gamification and peer influence elements, as well as a ‘virtual bank of positive travel behavior change transactions’ to provide support to financially disadvantaged older adults. More specifically, each time a person switches from driving to transit or from transit to walking, a credit is added to the virtual bank. Credits are accrued over time and used to provide access to transport services for financially disadvantaged people. We will do this by offering free transit passes, sponsored by the Challenge funds, to financially disadvantaged people as a ‘reward’ for participants using the platform and changing their travel behavior. This feature will encourage behavior change among some participants (motivated by altruistic goals).

We will undertake a free-living conditions pilot study with 40-50 older adults to evaluate the system’s capability to promote active travel and physical activity. Upon recruitment of the study participants, we will provide each person with a smartphone with the app installed. Our system will use the algorithms in the activity detection system and the signal from smartphone sensors to: (1) detect the participant’s travel mode; and (2) display in real-time the real location of the participant and of a ‘virtual person’ travelling on a realistic alternative travel mode (e.g., if the participant is travelling by motor vehicle, the ‘virtual person’ is travelling on transit or walking). The two locations will be displayed on a screen with two parallel lines showing the start and end of each journey (e.g., home and shops). At the end of the trip, the app will display ‘race statistics’, including total door-to-door travel time, time waiting (e.g., congestion and searching for parking), time in motion, steps walked, minutes of physical activity, calorie expenditure, emissions and energy consumption estimates, and cost (e.g., estimate of gasoline consumption or cost of transit ticket).

The participants will be given prompts prior to initiating trips (that are candidates for change), reminding them of the outcomes of their travel choices (in terms of time and other parameters). Participants will be ‘tracked’ using smartphone sensors and the activity detection system to determine the mode of travel used. The app will include gamification and peer influence elements, enabling participants to anonymously connect to the ‘community of participants’ and compete with themselves and each other (e.g., gain points for achieving personal goals or ‘out-competing’ other participants). In addition, our system will include the ‘virtual bank of positive transactions’ to encourage positive change motivated by altruistic goals.

The behavior change program will include the following stages: 1) data collection to establish current travel patterns, including mode of travel (e.g., car or walk), route, parking location (at destination), average walking/cycling speed and true door-to-door time (i.e., home to shopping mall pedestrian entrance) when travelling by motor vehicle/transit and when walking. 2) test the potential to influence behavior change simply by providing information about outcomes achieved by switching to walking or cycling. 3) test the potential to influence behavior change through a combination of information and peer influence/gamification. 4) test the potential to influence behavior change by including altruistic incentives via the ‘virtual bank of positive transactions’.

Why our solution will work?

Our team recently undertook three validation trials to determine the potential of personalized information to influence travel choices. In short, our validation trials were a resounding success. We now need to transfer this knowledge to key target beneficiaries (e.g., older adults) and scale the delivery of our innovative system.

The three pilot studies included over 120 participants in Melbourne (Australia) – two studies focused on shifting trips from motor vehicles to transit for commuting trips, and the other on shifting trips from transit to walking for the last segment of commuting trips. Participants were provided with a smartphone-based platform that displayed in real-time their location (on their routine mode of travel) and the location of a person undertaking the same trip on an alternative mode. The platform presented race statistics at the end of each journey, providing comparisons in terms of total door-to-door travel time (including the traditionally forgotten parking element), minutes ‘stuck in traffic’, minute of physical activity, calorie expenditure, emissions and cost. Participants were given prompts before initiating their journey ‘reminding’ them of the outcome of their travel decision. We tracked participants over a period of four weeks and achieved unprecedented levels of behavior change – 45%, 54% and 68% in the respective pilots. By comparison, multi-million-dollar and multi-year travel behavior change programs are considered successful when they achieve levels of change approaching 2-3%. Importantly, the change we achieved was revealed through the technology and not simply stated by users (as is the case for most trials to date).

Our team also undertook tests for short-, medium- and long-distance trips in inner Melbourne and middle suburban areas to determine the true door-to-door time of trips by motor vehicle (including the parking and walking components), on transit (including the walking segments at the beginning and end of the journey), by bike and on foot. The study revealed that there are a large proportion of trips (of a wide range of distances) for which transit is highly time competitive with the car. In addition, there are a large proportion of trips (up to around 3 miles in length) for which walking and cycling are highly competitive with all motorized modes. The travel time estimated using existing tools overestimated the walking, cycling and transit journey times (including the walking components) and underestimated the time when traveling by car (and completely ignored the parking and walking components at the end of the trip). This study demonstrates the need for providing real door-to-door time information (such as the one generated automatically by our system) to enable people to make informed decisions about their travel choices. This information alone has the potential to immediately make walking, cycling and transit more attractive and appealing than they appear to be using current tools.  

How does our solution address the opportunity area?

Our system, focused on promoting travel behavior change, is deliberately designed to make walking an attractive, accessible and enjoyable option for older adults. Our work to date around the world has revealed that (given congestion, parking constraints and operational issues associated with transit), walking is better (quicker) than all other travel modes for most trips in urban areas of up to 1-1.5 miles in length. Importantly, walking trips commence ‘immediately’ and take people precisely to the ultimate destination (unlike driving and transit). The primary outcome will be a system that provides real-time travel recommendations for older adults to walk more, while enabling them to (in all probability) arrive at their destinations even earlier (or at worst within a few minutes later than their current travel mode of choice).

Our system will promote an increase in physical activity for those people who change their behavior from driving to walking. This change will provide a tangible health benefit, while promoting a reduction in congestion and emissions associated with motor vehicle travel.

Pilot program

We propose to work closely with the City of Grand Rapids and the City of Tomorrow team to develop an appropriate pilot program as follows:

  1. Recruitment of program participants from a selected geographic and socioeconomic group of interest
  2. Funds to be allocated towards travel behavior change platform licensing costs (at a discounted rate), participant recruitment, outreach and credits for the virtual bank of positive transactions (to benefit disadvantaged older adults – potential beneficiaries to be selected with the City of Grand Rapids)
  3. Pilot period of three to six months

We have a strategic alliance with a Michigan-based transportation engineering firm that will support the program as appropriate.

The platform licensing for the City of Grand Rapids includes the following:

  1. A world-class travel behavior change intervention program that provides fully personalized active travel and physical activity recommendations for older adults that are commensurate with their individual capabilities and embedded into their daily travel routines.
  2. An adaptive Artificial Intelligence activity detection system
  3. A physical activity threshold estimator
  4. A peer influence, gamification and altruistic platform
  5. Cloud-based architecture and easy integration with third-party travel planning tools and health insurance incentive program apps

Describe who will use your solution (1,000 characters)

The target beneficiaries of this initiative are healthy older adults (65 years of age and older) in the City of Grand Rapids. For this particular Challenge, we will work with the City and the Challenge team to recruit 40-50 older adults from a suitable target population group (e.g., a selected neighborhood or community center). The inclusion criteria will be older adults who: 1) engage in daily trips on a regular basis; 2) self-report as healthy enough to walk at least 10-15 minutes; 3) live within 1 mile of shops, community center and/or other destinations in areas with no transit or within 3 miles in areas with frequent/high quality transit services; and 4) generally travel by car for a large proportion of trips (self-reported and to be validated with our activity detection system). Importantly, the framework/system to be piloted as part of this Challenge will be widely applicable and replicable with older adults across Grand Rapids, Michigan, the US and internationally.

Describe your solution's stage of development

  • Prototype - you have built a prototype and tested it with potential users

Tell us about your team or organization (500 characters)

Our team is composed of leaders in data science, AI, technology/software development, dynamic optimisation and medicine. For the Challenge, our team includes movendo (an Australian consultancy) and Harvard’s Motion Analysis Laboratory. movendo’s director, Jose Mantilla (M.Sc. Civil and Environmental Engineering, MIT), has over 20 years of international experience in government, industry and consulting. The Challenge will bring together organizations from distant places with a common goal.

Size of your team or organization

  • 2-10

Funding Request

  • $100,000

Describe how you would pilot your idea (1000 characters)

Challenge funds will be used to help cover the costs of running the proposed pilot, including travel behavior change platform licensing costs (at a discounted rate), participant recruitment and outreach. Challenge resources will also be used to fund the ‘virtual bank of positive travel behavior change transactions’ to benefit disadvantaged older adults. Importantly, the technology components are fully developed and ready to be deployed (after minor necessary adjustments to meet local conditions) – no Challenge funds will be used for technology development. A successful pilot will result in significant commercialization opportunities. The two primary channels will be: 1) technology licensing and/or intervention programs for the City of Grand Rapids after the pilot has been completed; and 2) technology licensing for health insurance/wellness organizations to enable telling people how to achieve physical activity goals (and associated incentives) as part of their daily travel routine.

Describe how you would measure the success of your pilot (1000 characters)

During the 3-6-month pilot, we will measure vehicle-miles-traveled (VMT), proportion of trips by travel mode and distance/steps/time walked will be the primary metrics. Our activity detection system automatically measures all three, enabling us to readily determine the pilot’s success. VMT changes will be combined with data about the vehicles used by each participant to estimate reductions in emissions/fuel consumption. VMT changes will also be used to estimate personalized reductions in road crash exposure. Increases in walking will be used to estimate personalized physical activity statistics (i.e., calorie expenditure and reductions in cardiovascular risks – using CDC data). The individual benefits will be aggregated to determine the pilot’s impact and estimate potential city-wide impacts. Surveys will be undertaken during/after the study to determine factors that motivate and/or inhibit travel behavior change, critical to expand the reach and replicability of the initiative.


Join the conversation:

Photo of Tom Bulten

Jose: Thank you for posting the proposal. I'm one of the facilitators on the forum. I'm intrigued by the ability of the proposed system to anticipate trips before they occur. I assume the AI learns from "daily trips on a regular basis" to predict future trips. For example, your test subjects would tend to travel to work every day at 8 am or tend to travel to the store every Wednesday at 2 pm. Is that correct? I'm also impressed by the "unprecedented levels of behavior change – 45%, 54% and 68% in the respective pilots."

Photo of Jose Mantilla

Hi Tom,
Thank you for your message.
As you correctly point out, our system learns a person's routine and identifies 'usual trips' (i.e., those that involve the same pairs of origins/destinations and occur at roughly the same times/days on a regular basis). Once the 'routine' has been established, the system 'anticipates' that a 'usual trip' is going to occur 'soon' and provides information to the user about the viable alternatives to the traditional mode of choice (when relevant).
In order to accomplish this last part (the recommendations for change), the system identifies all 'usual trips' and then determines those that are viable candidates for change (i.e., those trips for which switching to another mode or a combination of modes will result in marginal, if any, increases in the true 'door-to-door' travel time). For example, trips for which driving is clearly the only viable alternative (given the distances and transit services available), will be 'ignored'. On the other hand, those trips for which alternatives are time-competitive will be 'flagged' and alternatives will be provided to the user.
We are working with our scientific team to make the system truly adaptive and dynamic, so that it automatically adjusts travel mode/route recommendations based on (among other things) changes to personal (e.g., wearing high heels or traveling with a child) and environmental (e.g., longer than normal transit delays or rain) contexts.
With respect to the behavior change measured with our pilots, the results are extremely encouraging, as it suggests that truly personalized information alone can be a significant contributing factor to empowering people to make smarter/healthier/more sustainable travel choices.
I hope this information is useful. Please let me know if you have other questions.
Warm regards